{"build_at":"2026-10-02T18:09:57.287Z","entries":[{"type":"organization","slug":"abridge","href":"/companies/abridge/","name":"Abridge","summary":"Abridge develops AI software for health systems that records conversations between clinicians and patients and drafts clinical documentation from them for clinician review, with separate products for clinicians, nursing, and revenue-cycle teams. Its platform works inside electronic health record workflows.","keywords":"Abridge AI, Inc. United States (mailing address in Philadelphia, Pennsylvania) enterprise enterprise-software Abridge for clinicians Abridge for nursing"},{"type":"organization","slug":"adobe","href":"/companies/adobe/","name":"Adobe","summary":"Adobe makes creative, document, and marketing software, including Photoshop, Acrobat, and its enterprise content and marketing products. It develops Firefly, a family of generative AI models for images, video, and other media, which it offers in its apps and through APIs.","keywords":"Adobe Inc. San Jose, California enterprise enterprise-software consumer-products model-developer Adobe Firefly Adobe Firefly Services Acrobat AI Assistant"},{"type":"organization","slug":"agentic-ai-foundation","href":"/companies/agentic-ai-foundation/","name":"Agentic AI Foundation (AAIF)","summary":"The Agentic AI Foundation is a directed fund of The Linux Foundation that raises and spends funds to support open-source agentic AI projects approved by its Governing Board. The Linux Foundation announced it in December 2025 with three contributed projects: the Model Context Protocol from Anthropic, goose from Block, and AGENTS.md from OpenAI.","keywords":"agents open-source-steward The Linux Foundation"},{"type":"organization","slug":"agility-robotics","href":"/companies/agility-robotics/","name":"Agility Robotics","summary":"Agility Robotics builds Digit, a humanoid robot deployed commercially for warehouse, manufacturing, and distribution tasks such as tote handling, and Agility Arc, a cloud platform for managing Digit fleets. In September 2026 it announced Digit 5, a version designed to work close to people with less reliance on protective barriers.","keywords":"Agility Robotics, Inc. Salem, Oregon robotics robotics hardware-manufacturer Digit Agility Arc"},{"type":"organization","slug":"ai2","href":"/companies/ai2/","name":"Ai2 (Allen Institute for AI)","summary":"Ai2 is a non-profit AI research institute in Seattle, founded in 2014 by Paul Allen. Its work covers open models and training data, as well as robotics, conservation, and other applied research.","keywords":"The Allen Institute for Artificial Intelligence Seattle, Washington open-models research data model-developer research-lab nonprofit-research Ai2 Playground Asta Semantic Scholar"},{"type":"organization","slug":"all-hands-ai","href":"/companies/all-hands-ai/","name":"All Hands AI (OpenHands)","summary":"All Hands AI develops OpenHands, an open-source platform for building and running AI coding agents. It also offers a hosted service (OpenHands Cloud) and a commercial enterprise edition. The company now presents itself and its products under the OpenHands name.","keywords":"Massachusetts agents developer-platform enterprise-software OpenHands Cloud OpenHands Enterprise"},{"type":"organization","slug":"amazon","href":"/companies/amazon/","name":"Amazon (AWS)","summary":"Amazon runs retail, advertising, and cloud businesses and reports three segments: North America, International, and Amazon Web Services (AWS). Through AWS it offers AI services, including Amazon Bedrock for building generative AI applications and its own Amazon Nova models.","keywords":"Amazon.com, Inc. Seattle, Washington cloud frontier-models agents enterprise cloud-provider model-developer developer-platform consumer-products Amazon Bedrock Amazon Nova"},{"type":"organization","slug":"amd","href":"/companies/amd/","name":"AMD","summary":"AMD designs processors and accelerators, including EPYC server CPUs, AMD Instinct data center GPUs used for AI, adaptive SoCs and FPGAs, and networking products. It develops the ROCm open software stack for AI and HPC work on its GPUs.","keywords":"Advanced Micro Devices, Inc. Santa Clara, California chips chip-designer developer-platform AMD Instinct GPUs AMD ROCm"},{"type":"organization","slug":"anduril","href":"/companies/anduril/","name":"Anduril Industries","summary":"Anduril Industries describes itself as a defense technology company specializing in autonomous systems; it also develops the Lattice software platform. Its documented AI software includes Lattice for Command & Control, which it describes as an AI-enabled battle management platform, and Lattice for Mission Autonomy, which coordinates teams of uncrewed systems under a human operator.","keywords":"Anduril Industries, Inc. Costa Mesa, California defense robotics robotics hardware-manufacturer enterprise-software Lattice for Command & Control Lattice for Mission Autonomy Lattice SDK"},{"type":"organization","slug":"anthropic","href":"/companies/anthropic/","name":"Anthropic","summary":"Anthropic, which describes itself as an AI safety and research company, develops the Claude family of models. It offers Claude through web, desktop, and mobile apps, through the Claude Platform API, and through the Claude Code coding agent.","keywords":"Anthropic PBC San Francisco, California frontier-models agents research model-developer research-lab developer-platform consumer-products Claude Claude Platform (Claude API) Claude Code"},{"type":"organization","slug":"anyscale","href":"/companies/anyscale/","name":"Anyscale","summary":"Anyscale sells a managed platform for running Ray workloads (data processing, training, and online inference) across clouds and Kubernetes. Ray began at UC Berkeley's RISELab, and Anyscale contributed it to the PyTorch Foundation in 2025.","keywords":"Anyscale, Inc. San Francisco, California cloud developer-platform Anyscale Platform"},{"type":"organization","slug":"anysphere","href":"/companies/anysphere/","name":"Anysphere (Cursor)","summary":"Anysphere makes Cursor, an AI coding product with a desktop editor, coding agents, a command-line tool, and an automated code reviewer. Cursor offers models from several providers alongside its own Composer models. Since August 2026 Anysphere has been a wholly owned subsidiary of SpaceX.","keywords":"Anysphere, Inc. San Francisco, California agents developer-platform model-developer Cursor Cursor CLI Bugbot SpaceX"},{"type":"organization","slug":"apple","href":"/companies/apple/","name":"Apple","summary":"Apple makes and sells smartphones, personal computers, tablets, wearables, and accessories, and sells related services. Its AI features, called Apple Intelligence, run on-device and through Apple's Private Cloud Compute. Developers can reach Apple's on-device model through the Foundation Models framework, and Apple machine learning research publishes the open-source MLX array framework.","keywords":"Apple Inc. Cupertino, California research consumer-products hardware-manufacturer model-developer developer-platform Apple Intelligence Foundation Models framework"},{"type":"organization","slug":"apptronik","href":"/companies/apptronik/","name":"Apptronik","summary":"Apptronik develops the Apollo humanoid robot for manufacturing, logistics, and retail work. Its current model, Apollo 2, comes in bipedal and wheeled configurations; under a research partnership, data collected by Apollo 2 helps advance Google DeepMind's Gemini Robotics models.","keywords":"Apptronik, Inc. Austin, Texas robotics robotics hardware-manufacturer Apollo 2"},{"type":"organization","slug":"arc-institute","href":"/companies/arc-institute/","name":"Arc Institute","summary":"Arc Institute is an independent nonprofit research organization in Palo Alto, California, founded in 2021, that works in partnership with Stanford University, UC Berkeley, and UCSF. Its research spans experimental biology and computational models, including the Evo 2 DNA language model and virtual cell models.","keywords":"Arc Research Institute Palo Alto, California science research open-models research-lab nonprofit-research model-developer"},{"type":"organization","slug":"arc-prize-foundation","href":"/companies/arc-prize-foundation/","name":"ARC Prize Foundation","summary":"ARC Prize Foundation is a nonprofit that develops the ARC-AGI benchmarks, built from tasks that are simple for people but difficult for AI systems, and runs public competitions and model testing around them. It launched ARC-AGI-2 in March 2025 and the interactive ARC-AGI-3 in March 2026.","keywords":"ARC Prize Foundation (the name on its Form 990 for 2024); the website's terms of service and privacy policy name ARC Prize, Inc. at the same San Francisco address San Francisco, California research nonprofit-research open-source-steward ARC-AGI Toolkit (arc-agi Python package) ARC Prize 2026 (ARC-AGI-3 competition) ARC Prize Verified testing"},{"type":"organization","slug":"arcee-ai","href":"/companies/arcee-ai/","name":"Arcee AI","summary":"Arcee AI trains and publishes the Trinity family of open-weight language models, ranging from an on-device Nano model to a roughly 400-billion-parameter mixture-of-experts model. It also runs a hosted, OpenAI-compatible API.","keywords":"Arcee AI, Inc. San Francisco, California open-models research enterprise model-developer research-lab inference-provider Arcee Platform API Arcee chat"},{"type":"organization","slug":"arista-networks","href":"/companies/arista-networks/","name":"Arista Networks","summary":"Arista Networks makes network systems for data center, campus, and wide-area networks, built around its EOS network operating system. For AI workloads it sells the Etherlink family of Ethernet switches, including the 7800R AI spine, the 7060 AI leaf, and the 7700R4 Distributed Etherlink Switch, for networks that connect GPU and other accelerator clusters.","keywords":"Arista Networks, Inc. Santa Clara, California cloud enterprise hardware-manufacturer enterprise-software Arista Etherlink AI networking platforms Arista 7800R4 Series (AI Spine) Arista 7700R4 Distributed Etherlink Switch"},{"type":"organization","slug":"baseten","href":"/companies/baseten/","name":"Baseten","summary":"Baseten runs an inference platform for deploying open-source, custom, and fine-tuned models, offering per-token APIs for hosted open models, dedicated deployments in its cloud, a customer's cloud, or both, and managed training jobs. The company was started in 2019.","keywords":"Baseten Labs, Inc. San Francisco, California cloud inference-provider developer-platform Baseten Model APIs Baseten Dedicated Inference Baseten Training"},{"type":"organization","slug":"biohub","href":"/companies/biohub/","name":"Biohub (Chan Zuckerberg Biohub)","summary":"Biohub is a biomedical research organization, founded in 2016 and supported by the Chan Zuckerberg Initiative, that builds AI models of biology alongside data-generation, imaging, and laboratory technologies. After announcing in November 2025 that the EvolutionaryScale team would join it, Biohub now publishes the ESM protein models and runs the Biohub Platform for using them.","keywords":"Chan Zuckerberg Biohub, Inc. Redwood City, California science research research-lab nonprofit-research Biohub Platform"},{"type":"organization","slug":"block","href":"/companies/block/","name":"Block","summary":"Block is a publicly traded company that operates Square, a network for businesses, and Cash App, a consumer network, combining commerce, financial-services, software, and hardware products. Its Open Source Program Office created the goose AI agent, which Block contributed to the Agentic AI Foundation; it also describes AI features in its own products, including Managerbot for Square sellers and Moneybot in Cash App.","keywords":"Block, Inc. Oakland, California (principal executive office; no formal headquarters) agents consumer-products enterprise-software Managerbot Moneybot"},{"type":"organization","slug":"boltz","href":"/companies/boltz/","name":"Boltz","summary":"Boltz is a public benefit corporation that develops AI models for biomolecular structure prediction, binding-affinity estimation, and protein and small-molecule design. Its team started the Boltz models at MIT and founded Boltz PBC, announced in January 2026, which now presents Boltz-1, Boltz-2, and BoltzGen as its open-source models alongside a hosted platform and API.","keywords":"Boltz PBC Boston area, Massachusetts science open-models research model-developer research-lab developer-platform Boltz API Boltz Lab"},{"type":"organization","slug":"boston-dynamics","href":"/companies/boston-dynamics/","name":"Boston Dynamics","summary":"Boston Dynamics builds mobile robots: Spot, a quadruped used for industrial inspection; Stretch, a box-handling robot for warehouses; and Atlas, an electric humanoid being introduced with a select group of early adopters. Its Orbit software manages robot fleets and site inspection data. The company began in 1992 as a spin-off from the MIT Leg Lab and is controlled by Hyundai Motor Group.","keywords":"Boston Dynamics, Inc. Waltham, Massachusetts robotics robotics hardware-manufacturer Spot Atlas Stretch Orbit"},{"type":"organization","slug":"broadcom","href":"/companies/broadcom/","name":"Broadcom","summary":"Broadcom designs semiconductor products and sells infrastructure software, including VMware Cloud Foundation. Its AI-related semiconductor offerings include custom accelerators (XPUs), Ethernet switching and routing silicon, Ethernet NICs, and optical components, as well as racks and systems based on its XPUs.","keywords":"Broadcom Inc. Palo Alto, California chips enterprise chip-designer enterprise-software Tomahawk 6 (BCM78910 series) Ethernet switch VMware Private AI Foundation with NVIDIA"},{"type":"organization","slug":"carnegie-mellon-university","href":"/companies/carnegie-mellon-university/","name":"Carnegie Mellon University","summary":"Carnegie Mellon University is a university in Pittsburgh, Pennsylvania. It traces its origin to 1900, when Andrew Carnegie funded the Carnegie Technical Schools, and took its current form in 1967, when the Carnegie Institute of Technology merged with the Mellon Institute.","keywords":"Carnegie Mellon University Pittsburgh, Pennsylvania research university-lab research-lab"},{"type":"organization","slug":"cartesia","href":"/companies/cartesia/","name":"Cartesia","summary":"Cartesia develops speech models and offers them through an API, including the Sonic text-to-speech and Ink speech-to-text models, along with a platform for building and hosting voice agents. Its research centers on state space model (SSM) architectures, and it publishes some SSM language models on Hugging Face.","keywords":"Cartesia AI, Inc. San Francisco, California agents open-models model-developer developer-platform Sonic Ink Line"},{"type":"organization","slug":"center-for-ai-safety","href":"/companies/center-for-ai-safety/","name":"Center for AI Safety (CAIS)","summary":"The Center for AI Safety is a San Francisco-based research and field-building nonprofit whose stated mission is to reduce societal-scale risks from artificial intelligence. Its work covers technical safety research, including benchmarks, as well as field-building and advising industry and policymakers.","keywords":"Center for Artificial Intelligence Safety Inc San Francisco, California research nonprofit-research research-lab"},{"type":"organization","slug":"cerebras","href":"/companies/cerebras/","name":"Cerebras","summary":"Cerebras designs the Wafer-Scale Engine, a processor built from an entire silicon wafer, and rack-scale systems based on it for deployment in its own and customers' data centers. It also offers cloud inference and training services that run on its hardware.","keywords":"Cerebras Systems Inc. Sunnyvale, California chips cloud chip-designer hardware-manufacturer inference-provider compute-infrastructure Cerebras Inference Cerebras CS-4 system Cerebras Training Cloud"},{"type":"organization","slug":"chai-discovery","href":"/companies/chai-discovery/","name":"Chai Discovery","summary":"Chai Discovery develops AI models for predicting and designing molecular structures and interactions for drug discovery. It released Chai-1, a structure prediction model whose code and weights are now under Apache 2.0, and offers its later Chai-2 design models through an access program rather than as downloads.","keywords":"Chai Discovery, Inc. (as named in the original Chai-1 community license) San Francisco, California science open-models research model-developer research-lab Chai Lab web server Chai-2 access program"},{"type":"organization","slug":"character-ai","href":"/companies/character-ai/","name":"Character.AI","summary":"Character.AI runs a consumer chat app in which people talk by text or voice with AI characters created by users. It post-trains open models from other developers for its product, including CAI-Image, a family of image models post-trained from Qwen-Image.","keywords":"Character Technologies, Inc. Palo Alto, California consumer-products model-developer Character.AI"},{"type":"organization","slug":"chroma","href":"/companies/chroma/","name":"Chroma","summary":"Chroma develops the open-source Chroma database for vector, full-text, and metadata search, used for retrieval in AI applications, and sells Chroma Cloud, a hosted serverless version billed on usage.","keywords":"Chroma Inc. San Francisco, California data data-platform Chroma Cloud"},{"type":"organization","slug":"cisco","href":"/companies/cisco/","name":"Cisco","summary":"Cisco designs and sells networking, security, collaboration, and observability hardware and software. Its AI-related offerings include Silicon One networking silicon for AI data center networks, the Cisco Secure AI Factory with NVIDIA reference design, and Cisco AI Defense for securing AI applications and models.","keywords":"Cisco Systems, Inc. San Jose, California enterprise chips chip-designer enterprise-software Cisco Silicon One Cisco Secure AI Factory with NVIDIA Cisco AI Defense"},{"type":"organization","slug":"cline","href":"/companies/cline/","name":"Cline","summary":"Cline Bot Inc. develops Cline, an open-source AI coding agent offered as a VS Code extension, a command-line tool, an SDK, and a desktop app, with a separate JetBrains plugin. It also runs the Cline Provider, a hosted service that gives signed-in users credit-based access to models from other providers.","keywords":"Cline Bot Inc. United States (no headquarters city stated; the USPTO record gives a Wilmington, Delaware owner address, and the careers page lists on-site roles in San Francisco, California) agents developer-platform Cline for VS Code and JetBrains Cline CLI Cline Provider"},{"type":"organization","slug":"cloudflare","href":"/companies/cloudflare/","name":"Cloudflare","summary":"Cloudflare describes itself as a connectivity cloud company. Its AI offerings include Workers AI, which runs machine learning models on serverless GPUs across Cloudflare's global network, and AI Gateway, for monitoring and controlling an application's requests to AI model providers. It acquired the model-hosting platform Replicate in December 2025.","keywords":"Cloudflare, Inc. San Francisco, California cloud cloud-provider inference-provider developer-platform Workers AI AI Gateway Replicate"},{"type":"organization","slug":"cognition","href":"/companies/cognition/","name":"Cognition","summary":"Cognition makes Devin, an AI software engineering agent that runs in the cloud or on a developer's machine, together with Devin Desktop (formerly Windsurf) and Devin CLI. It also develops SWE-series coding models that are offered in these products.","keywords":"Cognition AI, Inc. San Francisco, California agents developer-platform model-developer Devin Devin Desktop (formerly Windsurf) Devin CLI"},{"type":"organization","slug":"common-crawl","href":"/companies/common-crawl/","name":"Common Crawl","summary":"Common Crawl is a nonprofit that crawls the web and makes its archives and derived datasets freely available. It has collected crawl data since 2008 and hosts it on Amazon Web Services through an AWS open data sponsorship program.","keywords":"The Common Crawl Foundation (listed by the IRS as Commoncrawl Foundation) Beverly Hills, California (address given in the Terms of Use and the IRS listing) data data-services"},{"type":"organization","slug":"contextual-ai","href":"/companies/contextual-ai/","name":"Contextual AI","summary":"Contextual AI builds an enterprise platform for creating AI agents and retrieval applications that work over an organization's own documents and data. It also sells component APIs for document parsing, instruction-following reranking, grounded generation, and natural-language unit tests for evaluating model outputs, and it has released open-weight reranker models.","keywords":"Contextual AI, Inc. Mountain View, California enterprise agents open-models model-developer enterprise-software developer-platform Contextual AI platform (Agent Composer) Contextual AI component APIs (Parse, Rerank, Generate, LMUnit)"},{"type":"organization","slug":"coreweave","href":"/companies/coreweave/","name":"CoreWeave","summary":"CoreWeave operates a cloud platform for AI workloads that combines GPU compute, managed Kubernetes on bare metal, storage, and networking for model training and inference. Its software offerings include the Weights & Biases developer platform.","keywords":"CoreWeave, Inc. Livingston, New Jersey cloud cloud-provider compute-infrastructure developer-platform CoreWeave GPU Compute CoreWeave Kubernetes Service (CKS) CoreWeave Forge (including Weights & Biases Models)"},{"type":"organization","slug":"cornell-university","href":"/companies/cornell-university/","name":"Cornell University","summary":"Cornell University is a privately endowed research university whose main campus is in Ithaca, New York; its charter was signed into law in 1865. Its AI-related research groups include the Kuleshov Group at Cornell Tech, which works on machine learning and generative models, and the Zhang Research Group in electrical and computer engineering, which studies hardware specialization for machine learning.","keywords":"Cornell University Ithaca, New York research university-lab research-lab"},{"type":"organization","slug":"crusoe","href":"/companies/crusoe/","name":"Crusoe","summary":"Crusoe builds and operates data centers for AI workloads and runs Crusoe Cloud, a GPU cloud platform with managed Kubernetes, Slurm, storage, and networking. It also offers managed inference and fine-tuning services for open models through Crusoe Intelligence Foundry.","keywords":"Denver, Colorado cloud cloud-provider compute-infrastructure inference-provider Crusoe Cloud Crusoe Managed Inference (Crusoe Intelligence Foundry) Crusoe AI data centers"},{"type":"organization","slug":"d-matrix","href":"/companies/d-matrix/","name":"d-Matrix","summary":"d-Matrix designs hardware and software for AI inference in data centers. Its products are the Corsair inference accelerator, the JetStream I/O accelerator for card-to-card communication, the Aviator software stack, and SquadRack rack-scale configurations that combine them.","keywords":"Santa Clara, California chips chip-designer hardware-manufacturer d-Matrix Corsair d-Matrix JetStream d-Matrix Aviator d-Matrix SquadRack"},{"type":"organization","slug":"databricks","href":"/companies/databricks/","name":"Databricks","summary":"Databricks sells a data and AI platform built on a lakehouse architecture that runs on AWS, Azure, and Google Cloud. The platform includes tools for building and serving AI agents and models, and Databricks-hosted access to open models through its Foundation Model APIs.","keywords":"San Francisco, California data enterprise agents data-platform enterprise-software inference-provider model-developer Databricks Data + AI Platform Agent Bricks Databricks Foundation Model APIs"},{"type":"organization","slug":"deep-cogito","href":"/companies/deep-cogito/","name":"Deep Cogito","summary":"Deep Cogito is a San Francisco AI company that post-trains open-weight language models, the Cogito series, on base models from other developers. It describes its training approach as Iterated Distillation and Amplification (IDA), a form of iterative self-improvement.","keywords":"Deep Cogito Inc. San Francisco, California open-models research model-developer research-lab"},{"type":"organization","slug":"dell-technologies","href":"/companies/dell-technologies/","name":"Dell Technologies","summary":"Dell Technologies sells PCs, servers, storage, and networking products and related services. Its Infrastructure Solutions Group includes a portfolio of AI-optimized servers, and it packages its AI infrastructure, data, and services offerings as the Dell AI Factory.","keywords":"Dell Technologies Inc. Round Rock, Texas enterprise hardware-manufacturer consumer-products Dell AI Factory Dell PowerEdge XE AI servers Dell AI Data Platform"},{"type":"organization","slug":"element-labs","href":"/companies/element-labs/","name":"Element Labs","summary":"Element Labs makes LM Studio, a desktop application for downloading and running language models on a local computer, with a chat interface and a local API server. It also publishes llmster, a headless version of LM Studio for servers, the lms command-line tool, and Python and TypeScript SDKs.","keywords":"Element Labs, Inc. New York, New York open-models developer-platform LM Studio llmster and the lms CLI LM Studio SDKs (lmstudio-python and lmstudio-js)"},{"type":"organization","slug":"eleutherai","href":"/companies/eleutherai/","name":"EleutherAI","summary":"EleutherAI is a non-profit AI research lab that trains and releases open models, datasets, and tools, and studies how models learn, how to evaluate them, and how to make open-weight systems safer. It began in July 2020 as a Discord community and incorporated as a non-profit research institute in early 2023.","keywords":"EleutherAI Institute (listed in the IRS exempt-organization extract as \"ELEUTHERAI INSTITUTE\") Washington, D.C. open-models research data research-lab nonprofit-research model-developer open-source-steward"},{"type":"organization","slug":"epoch-ai","href":"/companies/epoch-ai/","name":"Epoch AI","summary":"Epoch AI is an independent nonprofit that collects and analyzes data on AI development. It publishes data explorers and downloadable datasets on AI models, data centers, hardware, and related topics, runs benchmarks including FrontierMath, and publishes research reports, newsletters, and podcasts. It started in 2021 as a group of volunteers.","keywords":"Epoch Artificial Intelligence, Inc. San Francisco, California (address of record) research data research-lab nonprofit-research data-services Epoch AI data explorers and datasets"},{"type":"organization","slug":"essential-ai","href":"/companies/essential-ai/","name":"Essential AI","summary":"Essential AI is a San Francisco AI research company that publishes open models and data. Its releases include the Essential-Web v1.0 web dataset (June 2025) and the Rnj-1 language models for code and STEM tasks (December 2025).","keywords":"San Francisco, California open-models research data model-developer research-lab"},{"type":"organization","slug":"etched","href":"/companies/etched/","name":"Etched","summary":"Etched designs hardware for running AI model inference. Its website describes rack-scale \"frontier inference clusters\" for which it co-designs chips, packaging, interconnects, cooling, racks, and software, and says its first chip silicon was made on TSMC's N4P process.","keywords":"Etched, Inc. San Jose, California chips chip-designer hardware-manufacturer Etched rack-scale inference systems"},{"type":"organization","slug":"evolutionaryscale","href":"/companies/evolutionaryscale/","name":"EvolutionaryScale","summary":"EvolutionaryScale is a public benefit company that developed the ESM3 and ESM C (ESM Cambrian) protein language models. In November 2025 Biohub announced that the EvolutionaryScale team would join Biohub, and EvolutionaryScale's website now states that it is part of Biohub.","keywords":"EvolutionaryScale, PBC science open-models model-developer research-lab"},{"type":"organization","slug":"far-ai","href":"/companies/far-ai/","name":"FAR.AI","summary":"FAR.AI is a nonprofit that works on AI safety and security through in-house research, grantmaking, and events such as workshops and seminars. It also runs FAR.Labs, a coworking space in downtown Berkeley, California, for people and organizations working on trustworthy and secure AI.","keywords":"FAR AI, Inc. San Diego, California (address of record) research research-lab nonprofit-research"},{"type":"organization","slug":"figure","href":"/companies/figure/","name":"Figure","summary":"Figure builds general-purpose humanoid robots and Helix, its in-house vision-language-action model that controls them. Its current robot generation is Figure 03, designed for both home and commercial work.","keywords":"Figure AI Inc. San Jose, California robotics robotics hardware-manufacturer model-developer Figure 03 Helix"},{"type":"organization","slug":"fireworks-ai","href":"/companies/fireworks-ai/","name":"Fireworks AI","summary":"Fireworks AI operates a hosted platform for serving and training models, offering pay-per-token access to open models from other developers, dedicated deployments, and fine-tuning and reinforcement-learning training services.","keywords":"Fireworks.ai, Inc. San Mateo, California cloud inference-provider developer-platform Fireworks Inference Fireworks Training"},{"type":"organization","slug":"futurehouse","href":"/companies/futurehouse/","name":"FutureHouse","summary":"FutureHouse is a 501(c)(3) nonprofit research lab in San Francisco, founded in 2023, that builds AI systems intended to automate parts of scientific research, with a focus on biology. Its open releases include the PaperQA literature-agent package and the ether0 chemistry reasoning model.","keywords":"Future House Inc (as listed in the IRS exempt-organization extract) San Francisco, California science research agents open-models research-lab nonprofit-research model-developer open-source-steward"},{"type":"organization","slug":"genmo","href":"/companies/genmo/","name":"Genmo","summary":"Genmo is a San Francisco research company that builds video generation models, which it describes as world models. Its open release is Mochi 1, a text-to-video model published as a research preview in October 2024, and it runs a hosted playground for generating videos.","keywords":"Genmo Inc. San Francisco, California open-models research model-developer research-lab Genmo Playground"},{"type":"organization","slug":"georgia-institute-of-technology","href":"/companies/georgia-institute-of-technology/","name":"Georgia Institute of Technology","summary":"The Georgia Institute of Technology (Georgia Tech) is a public research university in Atlanta, Georgia, founded in 1885 and part of the University System of Georgia. Its AI research groups include the Polo Club of Data Science, which builds interactive tools for understanding machine learning models, among them Transformer Explainer.","keywords":"Atlanta, Georgia research university-lab research-lab"},{"type":"organization","slug":"github","href":"/companies/github/","name":"GitHub","summary":"GitHub runs a platform for hosting and collaborating on code, with cloud and on-premises offerings, and develops the GitHub Copilot AI coding assistant. It is owned by Microsoft, which reports GitHub within its Intelligent Cloud segment.","keywords":"GitHub, Inc. San Francisco, California agents developer-platform GitHub Copilot GitHub Copilot CLI Spec Kit Microsoft"},{"type":"organization","slug":"glean","href":"/companies/glean/","name":"Glean","summary":"Glean sells an enterprise AI platform that connects to a company's business applications and data sources to provide permission-aware search, an AI assistant, and AI agents. Its terms describe support for multiple third-party large language models.","keywords":"Glean Technologies, Inc. San Francisco, California enterprise agents enterprise-software developer-platform Glean Assistant Glean Agents Glean Developer Platform"},{"type":"organization","slug":"google","href":"/companies/google/","name":"Google (Alphabet)","summary":"Google is the largest of the businesses that make up Alphabet Inc. Alphabet reports Google through its Google Services and Google Cloud segments, and reports centralized research and development on frontier AI models as Alphabet-level activities. This record covers Google and its parent Alphabet together; Google DeepMind has its own record as a research unit.","keywords":"Alphabet Inc. (parent and SEC registrant); Google LLC (subsidiary) Mountain View, California frontier-models cloud open-models research model-developer research-lab cloud-provider developer-platform consumer-products Gemini API Gemini app Gemini Enterprise Agent Platform (formerly Vertex AI)"},{"type":"organization","slug":"google-deepmind","href":"/companies/google-deepmind/","name":"Google DeepMind","summary":"Google DeepMind is Google's AI research unit, formed in 2023 by combining DeepMind with the Brain team from Google Research. It is credited as the developer of the Gemma family of open-weight models.","keywords":"frontier-models open-models research research-lab model-developer Google (Alphabet)"},{"type":"organization","slug":"groq","href":"/companies/groq/","name":"Groq","summary":"Groq operates an inference cloud that runs AI models on LPU (Language Processing Unit) accelerators in data centers it deploys with partners. It developed the LPU and launched GroqCloud, and it offers developer access through its console as well as dedicated capacity tiers.","keywords":"United States (city not stated; legal mailing address is a P.O. box in Mountain View, California) cloud inference-provider compute-infrastructure GroqCloud"},{"type":"organization","slug":"harvard-university","href":"/companies/harvard-university/","name":"Harvard University","summary":"Harvard University is a university in Cambridge, Massachusetts, founded in 1636 by a vote of the Great and General Court of the Massachusetts Bay Colony. Its AI research includes the Kempner Institute for the Study of Natural and Artificial Intelligence, which studies the basis of intelligence in natural and artificial systems and publishes open-source software such as KempnerForge.","keywords":"President and Fellows of Harvard College Cambridge, Massachusetts research university-lab research-lab"},{"type":"organization","slug":"harvey","href":"/companies/harvey/","name":"Harvey","summary":"Harvey develops AI software for legal and professional services firms and in-house legal teams. Its platform includes agents that carry out multi-step legal tasks, Vault for storing and bulk-analyzing document collections, and Knowledge for legal, regulatory, and tax research across licensed and public sources.","keywords":"Harvey AI Corporation San Francisco, California enterprise agents enterprise-software Harvey agents Vault Knowledge"},{"type":"organization","slug":"hpe","href":"/companies/hpe/","name":"Hewlett Packard Enterprise (HPE)","summary":"HPE sells enterprise infrastructure and services spanning servers, hybrid cloud, and networking. Since November 2025 it has reported Cloud & AI and Networking as its main segments. Its AI offerings include HPE Private Cloud AI and other HPE AI Factory solutions, and HPE Cray supercomputers for HPC and AI workloads.","keywords":"Hewlett Packard Enterprise Company Spring, Texas enterprise hardware-manufacturer enterprise-software HPE Private Cloud AI HPE AI Factory HPE Cray Supercomputing"},{"type":"organization","slug":"hugging-face","href":"/companies/hugging-face/","name":"Hugging Face","summary":"Hugging Face runs the Hugging Face Hub, which hosts Git-based repositories of models, datasets, and demo apps (Spaces), and sells hosted inference and enterprise services. It also maintains open-source machine-learning libraries such as Transformers.","keywords":"Hugging Face, Inc. Brooklyn, New York open-models data developer-platform inference-provider open-source-steward Hugging Face Hub Inference Providers Inference Endpoints"},{"type":"organization","slug":"ibm","href":"/companies/ibm/","name":"IBM","summary":"IBM is a publicly traded technology company. In AI, it develops the Granite family of models and sells the watsonx products for building, running, and governing AI models and agents.","keywords":"International Business Machines Corporation Armonk, New York enterprise open-models model-developer enterprise-software IBM watsonx.ai IBM watsonx Orchestrate IBM watsonx.governance"},{"type":"organization","slug":"imbue","href":"/companies/imbue/","name":"Imbue","summary":"Imbue is a San Francisco company, founded in 2021, that builds software tools meant to keep AI agents under their users' control, with a focus on coding agents. Its products include Sculptor, a workspace for running coding agents in parallel; Vet, a tool that checks code changes and agent conversations for problems; and mngr, a command-line tool for managing coding agents. Most of its products are published as open source.","keywords":"Imbue, Inc. San Francisco, California agents research developer-platform research-lab Sculptor Vet mngr"},{"type":"organization","slug":"inception-labs","href":"/companies/inception-labs/","name":"Inception","summary":"Inception (Inception AI, Inc., at inceptionlabs.ai) develops Mercury, a family of diffusion language models that generate and refine many tokens in parallel rather than one at a time. It offers the models through an OpenAI-compatible API.","keywords":"Inception AI, Inc. Redwood City, California model-developer Inception API (Mercury models)"},{"type":"organization","slug":"inflection-ai","href":"/companies/inflection-ai/","name":"Inflection AI","summary":"Inflection AI develops its own foundation models and Pi, a conversational AI assistant it calls a personal intelligence partner. Its Inflection AI Labs group runs research and public product experiments, such as Pi Journeys, an assistant for life transitions like becoming a parent or changing careers.","keywords":"Inflection AI, Inc. San Francisco, California (mailing address) research model-developer consumer-products Pi"},{"type":"organization","slug":"intel","href":"/companies/intel/","name":"Intel","summary":"Intel designs and manufactures semiconductor products, including CPUs used in PCs and data centers. Its AI-related offerings include Intel Gaudi AI accelerators and the open-source OpenVINO inference toolkit.","keywords":"Intel Corporation Santa Clara, California chips chip-designer hardware-manufacturer developer-platform Intel Gaudi 3 AI accelerator OpenVINO toolkit"},{"type":"organization","slug":"lambda","href":"/companies/lambda/","name":"Lambda","summary":"Lambda is an AI cloud company that provides NVIDIA GPU infrastructure for training and inference, from on-demand instances and self-serve clusters to single-tenant superclusters. It was founded in 2012 by machine learning engineers.","keywords":"Lambda, Inc. San Jose, California cloud cloud-provider compute-infrastructure Lambda Instances 1-Click Clusters Lambda Superclusters"},{"type":"organization","slug":"lancedb","href":"/companies/lancedb/","name":"LanceDB","summary":"LanceDB builds data infrastructure for multimodal AI on the Lance columnar format: the open-source LanceDB embedded retrieval library and LanceDB Enterprise, a distributed multimodal lakehouse platform. In November 2025 LanceDB moved the Lance projects into a separate, community-governed lance-format GitHub organization.","keywords":"San Francisco, California data data-platform LanceDB Enterprise LanceDB OSS"},{"type":"organization","slug":"langchain","href":"/companies/langchain/","name":"LangChain","summary":"LangChain develops the open-source LangChain, LangGraph, and Deep Agents frameworks for building LLM agents. It also sells LangSmith, a commercial platform for tracing, evaluating, and deploying agents.","keywords":"LangChain Inc. San Francisco, California agents enterprise developer-platform enterprise-software LangSmith LangSmith Deployment LangSmith Fleet"},{"type":"organization","slug":"lawrence-berkeley-national-laboratory","href":"/companies/lawrence-berkeley-national-laboratory/","name":"Lawrence Berkeley National Laboratory","summary":"Lawrence Berkeley National Laboratory (Berkeley Lab) is a multiprogram national laboratory in Berkeley, California, managed by the University of California for the U.S. Department of Energy's Office of Science. It traces its founding to 1931. Berkeley Lab describes work on AI models and applications for science, AI-ready scientific datasets such as the Materials Project, and the computing and networking infrastructure used to train them.","keywords":"Berkeley, California research science research-lab"},{"type":"organization","slug":"lawrence-livermore-national-laboratory","href":"/companies/lawrence-livermore-national-laboratory/","name":"Lawrence Livermore National Laboratory","summary":"Lawrence Livermore National Laboratory (LLNL) is a U.S. Department of Energy national laboratory in Livermore, California, that began operations in 1952. It is operated by Lawrence Livermore National Security, LLC for DOE's National Nuclear Security Administration. LLNL describes AI as a core capability that it applies alongside experimental data, physics-based modeling, and high-performance computing, and it publishes the LBANN deep learning training toolkit as open-source software.","keywords":"Livermore, California research science defense research-lab"},{"type":"organization","slug":"lf-ai-data","href":"/companies/lf-ai-data/","name":"LF AI & Data Foundation","summary":"The LF AI & Data Foundation is a directed fund of The Linux Foundation that raises and spends funds to support open-source artificial intelligence, machine learning, and data projects. Its hosted projects include ONNX, which LF AI announced as a graduate-level project in November 2019.","keywords":"open-source-steward The Linux Foundation"},{"type":"organization","slug":"lightmatter","href":"/companies/lightmatter/","name":"Lightmatter","summary":"Lightmatter develops silicon photonics for AI infrastructure. Its Passage family covers optical interconnects in co-packaged, near-packaged, and on-board form factors, plus a photonic interposer reference platform, and its Guide light engine is a laser source for co-packaged optics. It says Passage and Guide are available to early-access partners.","keywords":"Lightmatter, Inc. Mountain View, California chips chip-designer hardware-manufacturer Passage L200 Passage L20 Passage M1000 evaluation kit Guide light engine"},{"type":"organization","slug":"lightning-ai","href":"/companies/lightning-ai/","name":"Lightning AI","summary":"Lightning AI created and maintains the PyTorch Lightning training framework and runs a cloud platform for building, training, and deploying AI models, including browser-based development environments (Studios) and on-demand GPUs. It merged with GPU infrastructure provider Voltage Park in January 2026 and operates under the Lightning AI name.","keywords":"San Francisco, California cloud cloud-provider compute-infrastructure developer-platform Lightning AI Studio Lightning GPU cloud"},{"type":"organization","slug":"linkedin","href":"/companies/linkedin/","name":"LinkedIn","summary":"LinkedIn is a professional networking service owned by Microsoft, which reports it within its Productivity and Business Processes segment. Its engineering team develops and maintains Liger Kernel, an open-source library of Triton GPU kernels for training large language models.","keywords":"LinkedIn Corporation Sunnyvale, California consumer-products enterprise-software Microsoft"},{"type":"organization","slug":"liquid-ai","href":"/companies/liquid-ai/","name":"Liquid AI","summary":"Liquid AI is a company spun out of MIT CSAIL in 2023 that develops Liquid Foundation Models (LFMs), hybrid models built for efficient deployment on phones, laptops, and other devices. It publishes LFM weights on Hugging Face and offers LEAP, a platform for deploying models on edge devices.","keywords":"Liquid AI, Inc. Cambridge, Massachusetts open-models research model-developer developer-platform LEAP (Liquid Edge AI Platform) Liquid Apollo"},{"type":"organization","slug":"llamaindex","href":"/companies/llamaindex/","name":"LlamaIndex","summary":"LlamaIndex develops the open-source LlamaIndex framework for building LLM agents and workflows over private data. It also sells LlamaParse, a hosted platform for document parsing, extraction, and indexing, which the company now describes as its main focus.","keywords":"LlamaIndex, Inc. San Francisco, California agents data enterprise developer-platform enterprise-software LlamaParse"},{"type":"organization","slug":"lmsys","href":"/companies/lmsys/","name":"LMSYS (Large Model Systems Organization)","summary":"LMSYS is a 501(c)(3) nonprofit that incubates open-source AI projects and research. It began in 2023 as a collaboration among UC Berkeley, Stanford, UC San Diego, Carnegie Mellon, and MBZUAI, and was incorporated as a nonprofit in September 2024. Its projects include the SGLang serving engine, FastChat, RouteLLM, and the Vicuna models.","keywords":"LMSYS Corp research open-models research-lab nonprofit-research open-source-steward"},{"type":"organization","slug":"luma-ai","href":"/companies/luma-ai/","name":"Luma AI","summary":"Luma AI develops generative models for video and images, including the Ray3.2 video model and the Uni-1.1 image model. It offers them through Luma Agents, a web-based creative workspace, and through a developer API.","keywords":"Luma AI, Inc. Redwood City, California model-developer consumer-products developer-platform Luma Agents Luma API"},{"type":"organization","slug":"marvell","href":"/companies/marvell/","name":"Marvell Technology","summary":"Marvell designs data infrastructure semiconductors for AI, cloud, carrier, and enterprise infrastructure. Its AI-related products include custom ASICs for cloud and AI data center customers, Teralynx Ethernet switch chips, optical interconnect DSPs, and the Photonic Fabric optical scale-up interconnect it gained by acquiring Celestial AI.","keywords":"Marvell Technology, Inc. Wilmington, Delaware chips chip-designer Marvell custom ASICs Teralynx Ethernet switches PAM4 optical DSPs Photonic Fabric technology"},{"type":"organization","slug":"mit","href":"/companies/mit/","name":"Massachusetts Institute of Technology","summary":"MIT is a university in Cambridge, Massachusetts, incorporated in 1861; its institutional policies describe it as independent and privately endowed. Its AI research groups include the MIT HAN Lab, which works on efficient AI methods such as model compression and quantization and maintains the AWQ code.","keywords":"Massachusetts Institute of Technology Cambridge, Massachusetts research university-lab research-lab"},{"type":"organization","slug":"meta","href":"/companies/meta/","name":"Meta","summary":"Meta operates social and messaging apps including Facebook, Instagram, WhatsApp, and Messenger, along with VR headsets and AI glasses. It develops AI models, including the Llama and Muse model families, and offers the Meta AI assistant across its apps.","keywords":"Meta Platforms, Inc. Menlo Park, California frontier-models open-models agents research model-developer research-lab consumer-products developer-platform Meta AI Meta Model API Muse (personal AI agent)"},{"type":"organization","slug":"metr","href":"/companies/metr/","name":"METR (Model Evaluation & Threat Research)","summary":"METR is a research nonprofit that evaluates frontier AI models, measuring their autonomous capabilities and assessing whether and when AI systems could pose catastrophic risks. It publishes research, evaluation reports on specific models, and open-source tooling for running evaluations.","keywords":"Model Evaluation and Threat Research, Inc. Berkeley, California research research-lab nonprofit-research Inspect-Hawk"},{"type":"organization","slug":"micron","href":"/companies/micron/","name":"Micron Technology","summary":"Micron makes memory and storage products, including DRAM, NAND, and NOR. Its Cloud Memory Business Unit covers memory for large hyperscale cloud customers and high-bandwidth memory (HBM) for all data center customers, and the company markets HBM, low-power server memory modules, and data center SSDs for AI workloads.","keywords":"Micron Technology, Inc. Boise, Idaho chips chip-designer hardware-manufacturer Micron HBM (HBM3E and HBM4) Micron SOCAMM2 Micron data center SSDs"},{"type":"organization","slug":"microsoft","href":"/companies/microsoft/","name":"Microsoft","summary":"Microsoft develops and sells software, cloud services, and devices. Its AI offerings include the Microsoft Foundry platform on Azure, Copilot assistants, and models developed in-house, including the Phi family of small language models and MAI models.","keywords":"Microsoft Corporation Redmond, Washington cloud enterprise open-models frontier-models research model-developer research-lab cloud-provider enterprise-software developer-platform consumer-products Microsoft Foundry Microsoft Copilot"},{"type":"organization","slug":"midjourney","href":"/companies/midjourney/","name":"Midjourney","summary":"Midjourney develops image-generation models and an image-to-video model and offers them through its subscription web service. It also publishes niji・journey, an anime-style image app whose accounts and plans can be synced with Midjourney accounts.","keywords":"Midjourney, Inc. South San Francisco, California model-developer consumer-products Midjourney niji・journey"},{"type":"organization","slug":"mintplex-labs","href":"/companies/mintplex-labs/","name":"Mintplex Labs","summary":"Mintplex Labs develops AnythingLLM, an MIT-licensed application for chatting with documents and running AI agents with local or hosted language models. It distributes a free desktop app and a self-hosted Docker version and operates AnythingLLM Cloud, a paid hosted service.","keywords":"Mintplex Labs, Inc. California agents consumer-products enterprise-software AnythingLLM Desktop AnythingLLM Cloud"},{"type":"organization","slug":"mlcommons","href":"/companies/mlcommons/","name":"MLCommons","summary":"MLCommons is an AI engineering consortium whose members include companies, academics, and nonprofits. It develops benchmarks, including the MLPerf performance suites and the AILuminate safety and security benchmarks, as well as open datasets and tools such as Croissant and MLCube. It launched in December 2020.","keywords":"MLCommons Association Dover, Delaware (address listed on the About page) data standards-body open-source-steward"},{"type":"organization","slug":"modal","href":"/companies/modal/","name":"Modal","summary":"Modal operates a cloud platform for AI and data workloads, including model inference, training and fine-tuning, batch jobs, and sandboxes for running untrusted code. Developers define workloads and hardware in Python code, and usage is billed per second.","keywords":"Modal Labs, Inc. New York, New York cloud cloud-provider inference-provider developer-platform Modal Inference Modal Sandboxes"},{"type":"organization","slug":"modular","href":"/companies/modular/","name":"Modular","summary":"Modular develops the Mojo programming language and MAX, a framework for serving and building AI models on CPUs, GPUs, and other accelerators, and operates Modular Cloud, a hosted inference service. Qualcomm completed its acquisition of Modular in July 2026.","keywords":"Modular Inc Los Altos, California cloud developer-platform inference-provider Modular Cloud MAX Mojo Qualcomm"},{"type":"organization","slug":"mozilla-ai","href":"/companies/mozilla-ai/","name":"Mozilla.ai","summary":"Mozilla.ai is a company backed by the Mozilla Foundation that builds open-source AI tools and products. Its projects include any-llm, a Python library for calling different model providers through one interface, and llamafile, which packages a model and its runtime into a single executable that runs locally.","keywords":"MZL.AI, PBC San Francisco, California open-models agents developer-platform open-source-steward llamafile any-llm"},{"type":"organization","slug":"nist","href":"/companies/nist/","name":"National Institute of Standards and Technology","summary":"The National Institute of Standards and Technology (NIST) is a U.S. federal agency founded in 1901 and now part of the Department of Commerce. It works on measurement science, standards, and technology; its AI work includes the AI Risk Management Framework, the Dioptra test platform, and a center for AI evaluation and voluntary guidelines.","keywords":"Gaithersburg, Maryland research standards-body research-lab"},{"type":"organization","slug":"new-york-university","href":"/companies/new-york-university/","name":"New York University","summary":"New York University is a private research university founded in 1831. It is anchored in New York City and has degree-granting campuses in Abu Dhabi and Shanghai.","keywords":"New York University New York, New York research university-lab research-lab"},{"type":"organization","slug":"nomic-ai","href":"/companies/nomic-ai/","name":"Nomic AI","summary":"Nomic is a New York City company that offers what it calls a domain-specific AI platform for architecture, engineering, and construction firms, including agents and an Agent API. It has also released open-source AI software and models, including the GPT4All desktop application and the Nomic Embed embedding models.","keywords":"Nomic, Inc. New York, New York enterprise agents open-models data model-developer enterprise-software developer-platform Nomic Platform Nomic Agent API Nomic text embedding API"},{"type":"organization","slug":"nous-research","href":"/companies/nous-research/","name":"Nous Research","summary":"Nous Research is an AI research company that trains and releases open-weight language models, most prominently the Hermes series of post-trained models, and builds open-source agent and training software such as Hermes Agent and the Atropos reinforcement-learning environment framework.","keywords":"Nous Research, Inc. United States (city not stated; the store privacy policy gives a contact address in Austin, Texas) open-models agents research model-developer research-lab developer-platform Nous Portal Hermes Agent"},{"type":"organization","slug":"nvidia","href":"/companies/nvidia/","name":"NVIDIA","summary":"NVIDIA designs accelerated computing platforms, including data center GPUs, networking, and complete systems, along with software such as CUDA for building and running AI workloads. Its annual report describes two segments, Compute & Networking and Graphics.","keywords":"NVIDIA Corporation Santa Clara, California chips open-models chip-designer model-developer developer-platform NVIDIA Vera Rubin platform NVIDIA DGX platform NVIDIA NIM NVIDIA CUDA Toolkit"},{"type":"organization","slug":"oak-ridge-national-laboratory","href":"/companies/oak-ridge-national-laboratory/","name":"Oak Ridge National Laboratory","summary":"Oak Ridge National Laboratory (ORNL) is a U.S. Department of Energy Office of Science national laboratory in Oak Ridge, Tennessee, managed by UT-Battelle LLC for DOE. It began in 1943 as part of the Manhattan Project. ORNL describes AI research embedded in scientific workflows and AI security research, and it maintains open-source AI software such as HydraGNN.","keywords":"Oak Ridge, Tennessee research science research-lab"},{"type":"organization","slug":"openai","href":"/companies/openai/","name":"OpenAI","summary":"OpenAI develops AI models and offers them through ChatGPT, the OpenAI API, and developer tools such as Codex. It has also published open-weight models, including the gpt-oss series released under the Apache 2.0 license.","keywords":"OpenAI Group PBC San Francisco, California frontier-models open-models agents model-developer research-lab developer-platform consumer-products ChatGPT OpenAI API Codex CLI"},{"type":"organization","slug":"oracle","href":"/companies/oracle/","name":"Oracle","summary":"Oracle sells database and enterprise software and operates Oracle Cloud, made up of Oracle Cloud Applications and Oracle Cloud Infrastructure (OCI). OCI compute includes GPU-based and bare-metal offerings for AI model training and inference, and OCI Generative AI is a managed service for building generative AI applications.","keywords":"Oracle Corporation Austin, Texas cloud enterprise data cloud-provider enterprise-software data-platform OCI AI Infrastructure OCI Generative AI"},{"type":"organization","slug":"pacific-northwest-national-laboratory","href":"/companies/pacific-northwest-national-laboratory/","name":"Pacific Northwest National Laboratory","summary":"Pacific Northwest National Laboratory (PNNL) is a U.S. Department of Energy national laboratory in Richland, Washington, managed and operated by Battelle for DOE. It traces its start to January 4, 1965, when the Atomic Energy Commission awarded Battelle the contract to operate the Hanford Laboratories. PNNL applies AI across discovery science, applied energy, and national security, and it maintains the open-source NeuroMANCER library.","keywords":"Richland, Washington research science research-lab"},{"type":"organization","slug":"palantir","href":"/companies/palantir/","name":"Palantir","summary":"Palantir builds software platforms for data integration, analytics, and operational decision-making used by government and commercial customers. Its Artificial Intelligence Platform (AIP) connects large language models with customer data and workflows inside its Foundry and Gotham platforms.","keywords":"Palantir Technologies Inc. Aventura, Florida enterprise data defense agents enterprise-software data-platform Palantir AIP (Artificial Intelligence Platform) Palantir Foundry Palantir Gotham"},{"type":"organization","slug":"perplexity","href":"/companies/perplexity/","name":"Perplexity","summary":"Perplexity makes an AI search and answer assistant that responds to questions with cited web sources. It also offers a developer API platform with web-grounded agent, search, embeddings, and model-routing APIs.","keywords":"Perplexity AI, Inc. San Francisco, California agents consumer-products developer-platform model-developer Perplexity Perplexity API Platform"},{"type":"organization","slug":"physical-intelligence","href":"/companies/physical-intelligence/","name":"Physical Intelligence","summary":"Physical Intelligence develops foundation models for robot control, including the π0, π0-FAST, and π0.5 vision-language-action models. It publishes code and checkpoints for these models in its openpi repository.","keywords":"San Francisco, California robotics open-models research model-developer research-lab robotics"},{"type":"organization","slug":"poolside","href":"/companies/poolside/","name":"Poolside","summary":"Poolside trains foundation models for software engineering and publishes the Laguna family of open-weight agentic coding models. It offers the models through an OpenAI-compatible API, a terminal coding agent, and self-managed inference deployments.","keywords":"Poolside, Inc. San Francisco, California open-models agents enterprise model-developer research-lab developer-platform Poolside API Poolside Agent CLI (pool) Self-managed Poolside inference"},{"type":"organization","slug":"prime-intellect","href":"/companies/prime-intellect/","name":"Prime Intellect","summary":"Prime Intellect provides GPU compute, a hosted post-training platform (Lab) with an Environments Hub for reinforcement-learning environments, inference, and code sandboxes, and trains open-weight models such as the INTELLECT series. It maintains the open-source prime-rl training framework and the verifiers environment library.","keywords":"Prime Intellect, Inc. Dover, Delaware (principal place of business listed in its SEC Form D) open-models cloud research model-developer research-lab compute-infrastructure developer-platform Prime Intellect Compute Lab (hosted training and Environments Hub) Prime Intellect Inference"},{"type":"organization","slug":"princeton-university","href":"/companies/princeton-university/","name":"Princeton University","summary":"Princeton University is a university in Princeton, New Jersey. Researchers at Princeton University and Princeton Language and Intelligence created SWE-bench, a benchmark of real-world GitHub issues for evaluating language models on software engineering tasks.","keywords":"Princeton, New Jersey research university-lab research-lab"},{"type":"organization","slug":"profluent","href":"/companies/profluent/","name":"Profluent","summary":"Profluent is a company in Emeryville, California that trains protein language models and uses them to design proteins, starting with gene editors. Its public releases include the ProGen3 generative models under a non-commercial weights license, the Profluent-E1 encoder models under a permissive license with attribution requirements, and the AI-designed OpenCRISPR-1 gene editor.","keywords":"Profluent Bio Inc. Emeryville, California science open-models research model-developer research-lab"},{"type":"organization","slug":"pytorch-foundation","href":"/companies/pytorch-foundation/","name":"PyTorch Foundation","summary":"The PyTorch Foundation is a directed fund of The Linux Foundation. It supports the PyTorch framework and other hosted projects, including vLLM, DeepSpeed, Ray, Helion, and Safetensors. It was formed in September 2022, when Meta moved PyTorch to the Linux Foundation.","keywords":"open-source-steward The Linux Foundation"},{"type":"organization","slug":"qualcomm","href":"/companies/qualcomm/","name":"Qualcomm","summary":"Qualcomm develops semiconductors and licenses its patent portfolio. Its annual report names on-device AI among its foundational technologies, including the Hexagon NPU in Snapdragon platforms. It also sells data center products, including rack-scale AI inference accelerators under the Qualcomm Dragonfly brand.","keywords":"QUALCOMM Incorporated San Diego, California chips chip-designer developer-platform Qualcomm Dragonfly AI accelerators (AI200, AI250, AI300) Qualcomm Hexagon NPU Qualcomm AI Hub"},{"type":"organization","slug":"red-hat","href":"/companies/red-hat/","name":"Red Hat","summary":"Red Hat is an enterprise open-source software company, owned by IBM since 2019, whose products include Red Hat Enterprise Linux and Red Hat OpenShift. Its Red Hat AI portfolio covers inference software built on vLLM and llm-d, OpenShift AI, and Red Hat Enterprise Linux AI. In 2025 it launched llm-d, an open-source distributed inference project.","keywords":"Red Hat, Inc. Raleigh, North Carolina enterprise cloud enterprise-software open-source-steward Red Hat AI Inference Red Hat OpenShift AI Red Hat Enterprise Linux AI IBM"},{"type":"organization","slug":"reflection-ai","href":"/companies/reflection-ai/","name":"Reflection AI","summary":"Reflection AI is an AI company that says it is building open-weight frontier models, along with open-source software and infrastructure for customizing and running them. As of this review it had not published any model weights.","keywords":"Reflection AI, Inc. New York, New York research model-developer research-lab"},{"type":"organization","slug":"reka","href":"/companies/reka/","name":"Reka","summary":"Reka is an AI research lab that develops multimodal models (the Spark, Edge, Flash, and Core lines) that process video, images, audio, and text, and offers them through an API. It also runs Reka Cloud, a platform for serverless inference, dedicated GPU capacity, and model fine-tuning, and it has released some models, including Reka Flash 3 and 3.1, as open weights.","keywords":"Reka AI, Inc. San Francisco, California open-models cloud research model-developer research-lab inference-provider Reka API Reka Cloud"},{"type":"organization","slug":"replit","href":"/companies/replit/","name":"Replit","summary":"Replit offers a hosted platform, available on the web, desktop, and mobile, where people describe software in natural language and create, publish, and run apps with the help of Replit Agent. It has also published two small code-completion models on Hugging Face.","keywords":"Replit, Inc. Foster City, California agents developer-platform model-developer Replit Replit Agent"},{"type":"organization","slug":"runway","href":"/companies/runway/","name":"Runway","summary":"Runway develops generative AI models, including the Gen-4.5 video model and world models such as GWM-1. It offers its own and some third-party models for video, image, and audio work through a creative web and mobile app and through a developer API.","keywords":"Runway AI, Inc. New York, New York model-developer consumer-products developer-platform Runway Runway API (Runway Dev)"},{"type":"organization","slug":"ssi","href":"/companies/ssi/","name":"Safe Superintelligence Inc.","summary":"Safe Superintelligence Inc. (SSI) is an AI research company that describes a single goal and product: building safe superintelligence. It says it is an American company with offices in Palo Alto and Tel Aviv.","keywords":"Safe Superintelligence Inc. Palo Alto, California research research-lab"},{"type":"organization","slug":"salesforce","href":"/companies/salesforce/","name":"Salesforce","summary":"Salesforce sells customer relationship management software and related cloud applications, which it now organizes on its AI-based Agentforce 360 Platform for building and running AI agents. Its research group, Salesforce AI Research, publishes models and datasets on Hugging Face.","keywords":"Salesforce, Inc. San Francisco, California enterprise agents research enterprise-software model-developer research-lab Agentforce Data 360 (formerly Data Cloud) Slack AI"},{"type":"organization","slug":"sambanova","href":"/companies/sambanova/","name":"SambaNova","summary":"SambaNova builds AI inference infrastructure: Reconfigurable Dataflow Unit (RDU) chips, SambaRack systems, the SambaStack platform for dedicated deployments, and the SambaCloud inference service for open models.","keywords":"SambaNova Systems, Inc. San Jose, California chips cloud chip-designer hardware-manufacturer inference-provider SambaCloud SambaStack SambaRack"},{"type":"organization","slug":"sandia-national-laboratories","href":"/companies/sandia-national-laboratories/","name":"Sandia National Laboratories","summary":"Sandia National Laboratories is a multimission U.S. Department of Energy laboratory with its headquarters in Albuquerque, New Mexico, and a second principal laboratory in Livermore, California. It is managed and operated for DOE's National Nuclear Security Administration by National Technology and Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc. Sandia lists secure AI, computing co-design, and scientific machine learning among its AI research areas.","keywords":"Albuquerque, New Mexico research science defense research-lab"},{"type":"organization","slug":"scale-ai","href":"/companies/scale-ai/","name":"Scale AI","summary":"Scale AI provides training data, annotation, and human-feedback services for AI model developers, along with model evaluation work and software for building AI agents. It sells to enterprises and to government customers, including national security agencies.","keywords":"San Francisco, California data enterprise agents defense data-services enterprise-software Scale Data Engine Scale GenAI Portfolio Scale Donovan"},{"type":"organization","slug":"servicenow","href":"/companies/servicenow/","name":"ServiceNow","summary":"ServiceNow sells the ServiceNow AI Platform, enterprise software for running business workflows, with generative AI tools (Now Assist), AI agents, and AI governance features built in. Its research teams publish the open-weight Apriel language models, and ServiceNow co-stewards the BigCode project with Hugging Face.","keywords":"ServiceNow, Inc. Santa Clara, California enterprise agents open-models enterprise-software model-developer research-lab ServiceNow AI Platform AI Control Tower"},{"type":"organization","slug":"sierra","href":"/companies/sierra/","name":"Sierra","summary":"Sierra provides a platform on which businesses build and run AI agents for customer service and customer experience, deployed across chat, messaging, email, and voice. It also maintains τ-bench, an open-source benchmark for evaluating customer-service agents.","keywords":"Sierra Technologies, Inc. San Francisco, California enterprise agents enterprise-software Agent Studio Ghostwriter Channels"},{"type":"organization","slug":"sifive","href":"/companies/sifive/","name":"SiFive","summary":"SiFive designs and licenses processor IP based on the open RISC-V instruction set. Its AI-related products include the SiFive Intelligence family of cores with vector processing and, in the XM series, matrix processing, and it offers the BigSky development server for porting data center software to RISC-V.","keywords":"SiFive, Inc. Santa Clara, California chips chip-designer SiFive Intelligence family (X100, X200, X300 series) SiFive Intelligence XM Series BigSky SF-2U870 datacenter development platform"},{"type":"organization","slug":"skild-ai","href":"/companies/skild-ai/","name":"Skild AI","summary":"Skild AI develops the Skild Brain, a robotics foundation model that it describes as able to control many kinds of robots, including quadrupeds, humanoids, tabletop arms, and mobile manipulators. In August 2026 it introduced S1, a manipulation model that learns tasks from video demonstrations.","keywords":"Pittsburgh, Pennsylvania robotics robotics model-developer Skild Brain S1"},{"type":"organization","slug":"snowflake","href":"/companies/snowflake/","name":"Snowflake","summary":"Snowflake operates a cloud data platform, which it calls the AI Data Cloud, deployed across multiple public clouds and regions. Its Cortex AI services run large language models from several providers inside the platform, and its research team released the Arctic language model with open weights in 2024.","keywords":"Snowflake Inc. Menlo Park, California (principal executive offices) data enterprise agents open-models data-platform inference-provider model-developer Snowflake Cortex AI Snowflake CoWork Snowflake AI Data Cloud"},{"type":"organization","slug":"spacex","href":"/companies/spacex/","name":"SpaceX","summary":"SpaceX reports three business segments: Space (reusable launch vehicles), Connectivity (the Starlink satellite network), and AI. SpaceX acquired xAI in February 2026, and its AI segment covers the Grok models, AI products for consumers and businesses, the X platform, and AI computing infrastructure.","keywords":"Space Exploration Technologies Corp. Starbase, Texas frontier-models model-developer compute-infrastructure consumer-products"},{"type":"organization","slug":"stanford-university","href":"/companies/stanford-university/","name":"Stanford University","summary":"Stanford University is a university in Stanford, California, founded in 1885 by Leland and Jane Stanford; it admitted its first students in 1891. Its AI research groups include the Center for Research on Foundation Models (CRFM) and the Stanford NLP Group.","keywords":"The Board of Trustees of the Leland Stanford Junior University Stanford, California research university-lab research-lab"},{"type":"organization","slug":"supermicro","href":"/companies/supermicro/","name":"Supermicro","summary":"Supermicro designs and builds servers, storage systems, and switches, and sells rack-scale IT solutions. Its annual report describes rack-scale systems for AI and high-performance computing and liquid cooling, offered through its Data Center Building Block Solutions.","keywords":"Super Micro Computer, Inc. San Jose, California enterprise hardware-manufacturer Supermicro GPU servers Data Center Building Block Solutions (DCBBS)"},{"type":"organization","slug":"tesla","href":"/companies/tesla/","name":"Tesla","summary":"Tesla makes electric vehicles and energy generation and storage products. Its annual report says the company is focused on bringing AI into the real world through products and services such as Full Self-Driving (Supervised) and Robotaxi, and on developing AI robots, including Optimus, which it describes as a general-purpose autonomous humanoid robot.","keywords":"Tesla, Inc. Austin, Texas robotics consumer-products hardware-manufacturer robotics Full Self-Driving (Supervised) Robotaxi"},{"type":"organization","slug":"linux-foundation","href":"/companies/linux-foundation/","name":"The Linux Foundation","summary":"The Linux Foundation is a nonprofit that hosts and supports open-source software, hardware, standards, and data projects. Among them is the PyTorch Foundation, which operates as one of its directed funds.","keywords":"The Linux Foundation San Francisco, California (legal postal address) open-source-steward"},{"type":"organization","slug":"university-of-texas-at-austin","href":"/companies/university-of-texas-at-austin/","name":"The University of Texas at Austin","summary":"The University of Texas at Austin is a public university in Austin, Texas, founded in 1883 and one of the institutions of The University of Texas System. Its AI research groups include the Robot Perception and Learning (RPL) Lab, which works on robot learning and embodied AI and publishes research code in its GitHub organization.","keywords":"Austin, Texas research university-lab research-lab"},{"type":"organization","slug":"thinking-machines-lab","href":"/companies/thinking-machines-lab/","name":"Thinking Machines Lab","summary":"Thinking Machines Lab is an AI research and product company. It runs Tinker, a hosted API for fine-tuning open-weight models, and in July 2026 released the open-weight models Inkling and Inkling-Small.","keywords":"Thinking Machines Lab, Inc. San Francisco, California open-models research model-developer research-lab developer-platform Tinker"},{"type":"organization","slug":"tiny-corp","href":"/companies/tiny-corp/","name":"tiny corp","summary":"The tiny corp writes and maintains tinygrad, an MIT-licensed neural network framework, and sells the tinybox, a multi-GPU computer for deep learning.","keywords":"tinygrad, Corp. San Diego, California hardware-manufacturer open-source-steward tinybox"},{"type":"organization","slug":"together-ai","href":"/companies/together-ai/","name":"Together AI","summary":"Together AI runs a cloud platform for inference, fine-tuning, and training on open-weight models from many developers, and rents GPU clusters. Its website and APIs are operated by Together Computer, Inc.","keywords":"Together Computer, Inc. San Francisco, California cloud open-models inference-provider compute-infrastructure developer-platform Together Serverless Inference Together Fine-Tuning Together GPU Clusters"},{"type":"organization","slug":"typesafe-ai","href":"/companies/typesafe-ai/","name":"TypeSafe AI","summary":"TypeSafe AI is a San Francisco company that develops what it calls System One models, AI models that answer typed questions about supplied input with structured values and probabilities rather than generated text. Its first such model, Jev, launched in early access in September 2026 through a hosted API.","keywords":"TypeSafe AI, Inc. San Francisco, California research enterprise model-developer developer-platform System One API (Jev) TypeSafe client SDKs"},{"type":"organization","slug":"university-of-california-san-diego","href":"/companies/university-of-california-san-diego/","name":"University of California San Diego","summary":"UC San Diego is a public university campus of the University of California, located in La Jolla, California, and founded in 1960. Its AI research groups include the Hao AI Lab, which develops and maintains open-source models, evaluations, and systems, including the FastVideo framework for video generation.","keywords":"La Jolla, California research university-lab research-lab"},{"type":"organization","slug":"uc-berkeley","href":"/companies/uc-berkeley/","name":"University of California, Berkeley","summary":"UC Berkeley is a public university and a campus of the University of California, which was founded in 1868 and moved to its Berkeley campus in 1873. Its AI-related research includes the Sky Computing Lab and the Gorilla project, which runs the Berkeley Function Calling Leaderboard.","keywords":"Berkeley, California research university-lab research-lab"},{"type":"organization","slug":"university-of-illinois-urbana-champaign","href":"/companies/university-of-illinois-urbana-champaign/","name":"University of Illinois Urbana-Champaign","summary":"The University of Illinois Urbana-Champaign is a public research university with a land-grant mission, founded in 1867 and governed by the Board of Trustees of the University of Illinois. Its AI research groups include U Lab, which maintains LLMRouter, an open-source library for routing requests among large language models.","keywords":"Urbana-Champaign, Illinois research university-lab research-lab"},{"type":"organization","slug":"university-of-washington","href":"/companies/university-of-washington/","name":"University of Washington","summary":"The University of Washington is a public university founded in 1861, with campuses in Seattle, Bothell, and Tacoma. Washington state law designates it as the state university established in Seattle.","keywords":"University of Washington Seattle, Washington research university-lab research-lab"},{"type":"organization","slug":"unsloth","href":"/companies/unsloth/","name":"Unsloth","summary":"Unsloth develops open-source software for fine-tuning, reinforcement learning, and running language, image, audio, and embedding models on local hardware, offered as a Python library (Unsloth Core), a web interface (Unsloth Studio), and a desktop app (Unsloth Desktop).","keywords":"Unsloth AI Inc. San Francisco, California (owner address in the company's USPTO trademark registration) open-models developer-platform Unsloth Desktop"},{"type":"organization","slug":"vercel","href":"/companies/vercel/","name":"Vercel","summary":"Vercel operates a cloud platform for building and deploying web applications and AI agents. Its AI offerings include the AI Gateway for calling models from many providers, the v0 app builder, and the open-source AI SDK for TypeScript.","keywords":"Vercel Inc. San Francisco, California cloud agents developer-platform cloud-provider Vercel Vercel AI Gateway v0 AI SDK"},{"type":"organization","slug":"waymo","href":"/companies/waymo/","name":"Waymo","summary":"Waymo develops the Waymo Driver, an autonomous driving system, and runs a public, fully autonomous ride-hailing service in U.S. cities. It began as Google's self-driving car project in 2009, became a separate company under Alphabet in 2016, and is reported by Alphabet within its Other Bets.","keywords":"Waymo LLC Mountain View, California robotics research robotics consumer-products research-lab Waymo (ride-hailing service) Waymo Driver Waymo for Business Google (Alphabet)"},{"type":"organization","slug":"weights-and-biases","href":"/companies/weights-and-biases/","name":"Weights & Biases","summary":"Weights & Biases develops an AI developer platform for tracking and comparing machine learning experiments (W&B Models) and for tracing and evaluating LLM applications and agents (W&B Weave), and maintains the open-source wandb Python library. CoreWeave acquired it in May 2025, and its products are now offered within CoreWeave Forge.","keywords":"Weights and Biases, LLC San Francisco, California agents enterprise developer-platform Weights & Biases Models W&B Weave CoreWeave"},{"type":"organization","slug":"world-labs","href":"/companies/world-labs/","name":"World Labs","summary":"World Labs develops world models that generate and reconstruct 3D environments. Its Marble web app and World API create explorable 3D worlds from text, images, video, or panoramas, and it maintains Spark, an open-source renderer for 3D Gaussian splats.","keywords":"World Labs Technologies, Inc. San Francisco, California research model-developer research-lab developer-platform Marble World API Spark"},{"type":"organization","slug":"writer","href":"/companies/writer/","name":"Writer","summary":"Writer is a San Francisco enterprise AI company, founded in 2020, whose platform includes the WRITER Agent, AI Studio, and related tools for business teams. The platform uses Writer's own Palmyra X models by default, and Writer has also published some Palmyra models as downloadable weights on Hugging Face.","keywords":"Writer, Inc. San Francisco, California enterprise agents open-models model-developer enterprise-software Palmyra models via the Writer API WRITER Agent"},{"type":"organization","slug":"xai","href":"/companies/xai/","name":"xAI (SpaceXAI)","summary":"xAI develops the Grok family of AI models and offers them through the Grok assistant and a developer API. Since February 2026 it has been a wholly owned subsidiary of SpaceX, whose AI segment also includes the X platform and AI computing infrastructure.","keywords":"Palo Alto, California frontier-models open-models model-developer consumer-products developer-platform compute-infrastructure Grok xAI API Grok Build SpaceX"},{"type":"organization","slug":"zyphra","href":"/companies/zyphra/","name":"Zyphra","summary":"Zyphra is a San Francisco AI company that trains open foundation models, including the Zamba hybrid state-space language models and the ZAYA1 mixture-of-experts models. It also runs Zyphra Cloud, a hosted platform for model inference and GPU compute.","keywords":"Zyphra Technologies, Inc. San Francisco, California open-models research cloud model-developer research-lab inference-provider Zyphra Cloud Maia"},{"type":"software","slug":"agent-development-kit","href":"/open/agent-development-kit/","name":"Agent Development Kit (ADK)","summary":"Agent Development Kit (ADK) is an open-source framework from Google for building, evaluating, and deploying AI agents, from single agents to multi-agent systems and graph-based workflows. Google publishes implementations for Python, Java, Go, TypeScript, and Kotlin. The README says ADK is optimized for Gemini but model-agnostic.","keywords":"framework agents multi-agent Google Apache-2.0 Google (Alphabet)"},{"type":"software","slug":"a2a-protocol","href":"/open/a2a-protocol/","name":"Agent2Agent (A2A) Protocol","summary":"Agent2Agent (A2A) is an open protocol for communication between AI agents built on different frameworks and run by different parties. Agents publish Agent Cards describing their capabilities and endpoints, and exchange messages and long-running tasks over JSON-RPC, gRPC, or HTTP+JSON bindings without exposing internal state or tools. The project publishes the specification, SDKs, and samples.","keywords":"framework agents protocol interoperability Agent2Agent project (Linux Foundation; Agentic AI Foundation project) Apache-2.0 Agentic AI Foundation (AAIF) The Linux Foundation Google (Alphabet)"},{"type":"software","slug":"agents-md","href":"/open/agents-md/","name":"AGENTS.md","summary":"AGENTS.md is an open convention for a Markdown file, placed in a code repository, that gives AI coding agents project-specific context and instructions such as build steps, test commands, and code conventions. It is plain Markdown with no required fields; in a monorepo, nested files can apply to subprojects, with the file closest to the edited code taking precedence. OpenAI released it in August 2025 and contributed it to the Agentic AI Foundation in December 2025.","keywords":"framework agents coding-agent documentation AGENTS.md a Series of LF Projects, LLC (Agentic AI Foundation project) MIT Agentic AI Foundation (AAIF) The Linux Foundation OpenAI"},{"type":"software","slug":"ai-sdk","href":"/open/ai-sdk/","name":"AI SDK","summary":"The AI SDK is an open-source TypeScript toolkit from Vercel for building AI applications and agents. AI SDK Core gives one API for generating text and structured output, calling tools, and building agents across model providers, and AI SDK UI provides hooks for chat and generative interfaces in web frameworks.","keywords":"framework agents llm-applications developer-tools Vercel Apache-2.0 Vercel"},{"type":"evaluation-tool","slug":"ailuminate","href":"/open/ailuminate/","name":"AILuminate","summary":"AILuminate is MLCommons' family of safety and security benchmarks for generative AI systems, organized around 12 hazard categories. The v1.0 safety benchmark (December 2024) tests general-purpose chat systems with more than 24,000 human-written prompts and grades responses with an ensemble of evaluator models; a jailbreak benchmark measures how safety degrades under attack.","keywords":"Safety v1.0 and v1.1; Jailbreak v0.5 eval benchmark evaluation MLCommons (AI Risk & Reliability working group) CC-BY-4.0 Apache-2.0 MLCommons"},{"type":"software","slug":"anythingllm","href":"/open/anythingllm/","name":"AnythingLLM","summary":"AnythingLLM is an application for chatting with documents and building AI agents on top of local or hosted language models. It is available as a desktop app for macOS, Windows, and Linux and as a multi-user Docker version, with connectors to many model providers and vector databases.","keywords":"runtime local-inference agents retrieval Mintplex Labs MIT Mintplex Labs"},{"type":"model-family","slug":"apriel","href":"/open/apriel/","name":"Apriel","summary":"Apriel is ServiceNow's series of small open-weight language models, built by its SLAM (ServiceNow Language Models) lab. Apriel-5B was the first release. It was followed by 15B-parameter reasoning models: Apriel-Nemotron-15b-Thinker, the multimodal Apriel-1.5-15b-Thinker, and Apriel-1.6-15b-Thinker (December 2025). In April 2026 ServiceNow released Super Apriel, a 15B model derived from Apriel 1.6 whose layers can switch between attention and linear-time mixers at inference time.","keywords":"model language-model reasoning multimodal ServiceNow (SLAM lab) ServiceNow"},{"type":"model-release","slug":"apriel-1-6-15b-thinker","href":"/open/apriel-1-6-15b-thinker/","name":"Apriel-1.6-15B-Thinker","summary":"A multimodal reasoning model of about 15B parameters that accepts text and images and writes its reasoning steps before a marked final answer. ServiceNow released it in December 2025 as an update to Apriel-1.5-15b-Thinker, with further continual pretraining, supervised fine-tuning, and reinforcement learning aimed partly at shorter reasoning.","keywords":"Apriel-1.6-15b-Thinker model language-model reasoning multimodal vision-language ServiceNow (SLAM lab) MIT ServiceNow Apriel"},{"type":"evaluation-tool","slug":"arc-agi","href":"/open/arc-agi/","name":"ARC-AGI","summary":"ARC-AGI is a series of benchmarks that measure how efficiently a system acquires new skills on unfamiliar tasks, using tasks that rely only on basic \"core knowledge\" priors and are easy for people. ARC-AGI-1 (introduced in 2019) and ARC-AGI-2 (March 2025) are grid puzzles in which the solver infers a transformation from a few input-output examples; ARC-AGI-3 (March 2026) consists of interactive, turn-based game environments in which agents must explore and infer goals without explicit instructions. Public tasks are released for each version, while semi-private and private evaluation sets used for official scoring are held back.","keywords":"eval evaluation benchmark reasoning ARC Prize Foundation Apache-2.0 Apache-2.0 MIT ARC Prize Foundation"},{"type":"software","slug":"autogen","href":"/open/autogen/","name":"AutoGen","summary":"AutoGen is Microsoft's open-source Python framework for building multi-agent AI applications that act autonomously or alongside people. It comprises an event-driven Core, the higher-level AgentChat API, Extensions for external services such as MCP servers and code execution, and the no-code AutoGen Studio. The project is now in maintenance mode, and Microsoft directs new users to its successor, Microsoft Agent Framework.","keywords":"framework agents Microsoft MIT CC-BY-4.0 Microsoft"},{"type":"software","slug":"awq","href":"/open/awq/","name":"AWQ (Activation-aware Weight Quantization)","summary":"AWQ is a post-training method and codebase for low-bit (INT3/INT4) weight-only quantization of large language models, including instruction-tuned and multimodal models. It uses activation statistics to find the most important weight channels and scales them before quantization, without backpropagation. The repository also includes 4-bit CUDA kernels and TinyChat, an inference interface for running quantized models on desktop and edge GPUs.","keywords":"framework quantization inference compression MIT HAN Lab MIT Massachusetts Institute of Technology"},{"type":"evaluation-tool","slug":"berkeley-function-calling-leaderboard","href":"/open/berkeley-function-calling-leaderboard/","name":"Berkeley Function Calling Leaderboard (BFCL)","summary":"The Berkeley Function Calling Leaderboard (BFCL) is a benchmark and live leaderboard from UC Berkeley's Gorilla project for evaluating how well language models call functions and tools. It checks serial and parallel function calls in several programming languages using abstract syntax tree (AST) evaluation. Later versions add multi-turn and multi-step scenarios and, in V4, agentic tasks such as web search, memory management, and format sensitivity.","keywords":"2026.3.23 eval evaluation benchmark agents Gorilla project, UC Berkeley Apache-2.0 Apache-2.0 University of California, Berkeley"},{"type":"evaluation-tool","slug":"bigcodebench","href":"/open/bigcodebench/","name":"BigCodeBench","summary":"BigCodeBench is a Python code-generation benchmark from the BigCode project with 1,140 function-level tasks that require calls to 139 libraries across 7 domains. It has two variants, Complete (prompts with structured docstrings) and Instruct (natural-language instructions for chat models), plus a 148-task Hard subset, and it is scored by executing unit tests. The GitHub repository was archived and made read-only on July 20, 2026.","keywords":"eval evaluation benchmark code-model BigCode project (open scientific collaboration) Hugging Face (co-leads the BigCode steering committee) ServiceNow (co-leads the BigCode steering committee) Apache-2.0 Apache-2.0 Hugging Face ServiceNow"},{"type":"software","slug":"bitnet","href":"/open/bitnet/","name":"bitnet.cpp (BitNet inference framework)","summary":"bitnet.cpp is Microsoft's official inference framework for 1-bit large language models, such as BitNet b1.58, whose weights are ternary (1.58-bit). It provides optimized kernels for running these models on x86 and ARM CPUs, with a separate GPU inference kernel, and is based on llama.cpp.","keywords":"runtime inference local-inference quantization Microsoft MIT Microsoft"},{"type":"model-family","slug":"boltz-2","href":"/open/boltz-2/","name":"Boltz-2","summary":"Boltz-2 is a biomolecular foundation model that predicts the 3D structures of complexes of proteins, nucleic acids, and small molecules together with protein-ligand binding affinity. It was released in June 2025 by the Boltz team at MIT Jameel Clinic with Recursion and is now presented by Boltz PBC as one of its open-source models. It ships as a structure and confidence checkpoint plus a separate affinity checkpoint.","keywords":"model science structure-prediction Boltz (Boltz PBC) Boltz Massachusetts Institute of Technology"},{"type":"model-release","slug":"boltz-2-v2","href":"/open/boltz-2-v2/","name":"Boltz-2 (boltz2_conf and boltz2_aff checkpoints)","summary":"The Boltz-2 weights published on June 6, 2025: a structure and confidence checkpoint (boltz2_conf.ckpt) and a separate binding-affinity checkpoint (boltz2_aff.ckpt), released together in one Hugging Face repository and loaded by the boltz package, which runs the latest model by default. The two checkpoints are assessed together because the affinity prediction runs as part of the same Boltz-2 pipeline.","keywords":"boltz-community/boltz-2: boltz2_conf.ckpt + boltz2_aff.ckpt model science structure-prediction Boltz (Boltz PBC) MIT Boltz Massachusetts Institute of Technology Boltz-2"},{"type":"model-family","slug":"chai-1","href":"/open/chai-1/","name":"Chai-1","summary":"Chai-1 is a multi-modal foundation model from Chai Discovery for predicting molecular structures, covering proteins, small molecules, DNA, RNA, glycosylations, and other modifications. It was first released in September 2024, and its license changed in November 2024 (see the release record). The chai_lab package (latest PyPI release 0.6.1, March 2025) runs it locally.","keywords":"model science structure-prediction Chai Discovery Chai Discovery"},{"type":"model-release","slug":"chai-1-v0-6","href":"/open/chai-1-v0-6/","name":"Chai-1 (weights used by chai_lab 0.6.1)","summary":"The Chai-1 weights used by the current chai_lab package (version 0.6.1, published March 18, 2025). They are distributed as six exported PyTorch module files (feature embedding, bond projection, token embedder, trunk, diffusion module, and confidence head). The package fetches them from Chai's asset server, and the same six files are in the Hugging Face repository, last updated February 18, 2025. The sources reviewed do not say which earlier package versions used these same files.","keywords":"chai_lab 0.6.1 (models_v2 components) model science structure-prediction Chai Discovery Apache-2.0 Chai Discovery Chai-1"},{"type":"software","slug":"chroma-db","href":"/open/chroma-db/","name":"Chroma (open-source database)","summary":"Chroma is an open-source database for AI applications that stores documents, embeddings, and metadata and supports vector, full-text, regex, and metadata search. It can run in memory inside a program, persist to local disk, or run as a server, with client libraries for Python, JavaScript/TypeScript, and Rust.","keywords":"runtime vector-database retrieval embedding Chroma Apache-2.0 Chroma"},{"type":"model-family","slug":"chronos","href":"/open/chronos/","name":"Chronos (time series models)","summary":"Chronos is Amazon's family of pretrained time-series forecasting models. It includes the original T5-based Chronos models (March 2024, 8M to 710M parameters), the patch-based Chronos-Bolt models (November 2024), and Chronos-2 (October 2025), a 120M-parameter encoder-only model that adds multivariate and covariate-informed forecasting.","keywords":"model time-series forecasting Amazon (Amazon Web Services) Amazon (AWS)"},{"type":"model-release","slug":"chronos-2","href":"/open/chronos-2/","name":"Chronos-2","summary":"Chronos-2 is a 120M-parameter, encoder-only time-series foundation model from Amazon, released on October 20, 2025. Inspired by the T5 encoder, it produces multi-step quantile forecasts and uses group attention to learn in context across related series and covariates, supporting univariate, multivariate, and covariate-informed forecasting with contexts up to 8,192 steps and horizons up to 1,024 steps.","keywords":"Chronos-2 (amazon/chronos-2, 120M) model time-series forecasting Amazon (Amazon Web Services) Apache-2.0 Apache-2.0 Amazon (AWS) Chronos (time series models)"},{"type":"software","slug":"cline-extension","href":"/open/cline-extension/","name":"Cline (open-source coding agent)","summary":"Cline is an open-source AI coding agent that reads a project, edits files, and runs terminal commands, asking for approval at each step unless auto-approve is turned on. Its public repository holds a shared agent SDK, a command-line tool, the VS Code extension, and a desktop app for macOS and Windows; the JetBrains plugin is not open-sourced.","keywords":"runtime agents coding-agent developer-tools Cline Bot Inc. Apache-2.0 Cline"},{"type":"model-family","slug":"clip","href":"/open/clip/","name":"CLIP","summary":"CLIP (Contrastive Language-Image Pre-Training) is an OpenAI family of paired image and text encoders trained with a contrastive objective on 400 million image-text pairs, which lets the model classify images zero-shot from natural-language labels. OpenAI released the models in stages from January 2021 to April 2022, from ResNet-50 and ViT-B/32 up to ViT-L/14 and ViT-L/14 at 336-pixel resolution.","keywords":"model multimodal vision-language embedding OpenAI OpenAI"},{"type":"model-release","slug":"clip-vit-large-patch14","href":"/open/clip-vit-large-patch14/","name":"CLIP ViT-L/14","summary":"CLIP ViT-L/14 pairs a ViT-L/14 Vision Transformer image encoder with a masked self-attention Transformer text encoder, trained contrastively to match images with their captions. OpenAI released it in January 2022; a variant further trained at 336-pixel resolution (ViT-L/14@336px) followed in April 2022.","keywords":"ViT-L/14 model multimodal vision-language embedding OpenAI MIT OpenAI CLIP"},{"type":"software","slug":"codex-cli","href":"/open/codex-cli/","name":"Codex CLI","summary":"Codex CLI is OpenAI's open-source coding agent that runs locally in a terminal, where it can inspect, edit, and run code in a local repository. The README points to separate Codex experiences for code editors, a desktop app, and Codex Web, OpenAI's cloud-based agent.","keywords":"runtime agents coding-agent developer-tools OpenAI Apache-2.0 OpenAI"},{"type":"model-family","slug":"cogito","href":"/open/cogito/","name":"Cogito","summary":"Cogito is Deep Cogito's series of open-weight language models; the v2 and v2.1 models are hybrid reasoning models that can answer directly or reason first. Cogito v1 Preview (April 2025) covered 3B to 70B models on Llama and Qwen bases; Cogito v2 Preview (July 2025) added 70B dense, 109B MoE, 405B dense, and 671B MoE models on Llama and DeepSeek bases; and Cogito v2.1 (November 2025) is a 671B MoE model with 37B active parameters on a DeepSeek base.","keywords":"model language-model reasoning Deep Cogito Deep Cogito"},{"type":"model-release","slug":"cogito-v2-preview-llama-405b","href":"/open/cogito-v2-preview-llama-405b/","name":"Cogito v2 Preview Llama 405B","summary":"A 405B-parameter dense hybrid reasoning model that Deep Cogito post-trained from Meta's Llama 3.1 405B, released in July 2025 as one of four Cogito v2 Preview models. It can answer directly or reason first, supports a 128k-token context, and was trained in over 30 languages using Deep Cogito's Iterated Distillation and Amplification approach.","keywords":"cogito-v2-preview-llama-405B model language-model reasoning Deep Cogito Llama 3.1 Community License Agreement Deep Cogito Cogito"},{"type":"model-release","slug":"cogito-v2-1-671b","href":"/open/cogito-v2-1-671b/","name":"Cogito v2.1 671B","summary":"A 671B-parameter mixture-of-experts model with 37B active parameters that Deep Cogito post-trained in-house from the DeepSeek-V3 base model. It is a hybrid reasoning model that can answer directly or reason first, supports a 128k-token context, and was trained in over 30 languages; Deep Cogito says process supervision of reasoning chains lets it reach answers with shorter reasoning.","keywords":"cogito-671b-v2.1 model language-model reasoning mixture-of-experts Deep Cogito MIT Deep Cogito Cogito"},{"type":"dataset","slug":"common-crawl-corpus","href":"/open/common-crawl-corpus/","name":"Common Crawl corpus","summary":"An archive of web crawl data collected by Common Crawl's crawlers since 2008, released as periodic crawls. Each crawl is published as WARC files (raw HTTP responses, requests, and metadata), WAT files (computed metadata as JSON), and WET files (extracted plain text), together with URL indexes.","keywords":"CC-MAIN-2026-39 (latest crawl at review) dataset pretraining-data Common Crawl Foundation Common Crawl Terms of Use (limited, non-exclusive license) Common Crawl"},{"type":"model-family","slug":"contextual-reranker","href":"/open/contextual-reranker/","name":"Contextual AI Reranker","summary":"Contextual AI's reranker models score how relevant retrieved documents are to a query and can follow a natural-language instruction about how to rank them, for example to prefer recent sources. Reranker v2, released in August 2025, comes in 1B, 2B, and 6B sizes with quantized versions, supports more than 100 languages and inputs of up to 32K tokens, and is published as open weights for noncommercial use; managed versions are served through Contextual AI's Rerank API.","keywords":"model retrieval reranker Contextual AI Contextual AI"},{"type":"model-release","slug":"contextual-reranker-v2-2b","href":"/open/contextual-reranker-v2-2b/","name":"Contextual AI Reranker v2 2B","summary":"The 2B-parameter size of Contextual AI Reranker v2, an instruction-following multilingual reranker released in August 2025. Given a query, an optional natural-language instruction, and a document, it outputs a relevance score; the card lists support for more than 100 languages and inputs of up to 32K tokens.","keywords":"ctxl-rerank-v2-instruct-multilingual-2b model retrieval reranker Contextual AI CC-BY-NC-SA-4.0 Contextual AI Contextual AI Reranker"},{"type":"software","slug":"coremltools","href":"/open/coremltools/","name":"Core ML Tools (coremltools)","summary":"Core ML Tools (coremltools) is Apple's Python package for converting machine learning models from libraries such as PyTorch, TensorFlow, scikit-learn, XGBoost, and LibSVM into the Core ML format, and for reading, writing, and optimizing Core ML models. On macOS it can also check a conversion by running predictions through Core ML.","keywords":"framework model-conversion inference local-inference Apple BSD-3-Clause Apple"},{"type":"software","slug":"deepspeed","href":"/open/deepspeed/","name":"DeepSpeed","summary":"DeepSpeed is an open-source PyTorch library for distributed training and inference of large models; its features include the ZeRO memory optimizations, 3D parallelism, Ulysses sequence parallelism, and mixture-of-experts support. Microsoft contributed it to LF AI & Data as an incubation project in February 2025, and in May 2025 it became a hosted project of the PyTorch Foundation, which names Microsoft as the contributor.","keywords":"framework training distributed-computing inference PyTorch Foundation (hosted project) DeepSpeed Technical Steering Committee (project committers) Apache-2.0 CC-BY-4.0 PyTorch Foundation The Linux Foundation Microsoft"},{"type":"software","slug":"diffusers","href":"/open/diffusers/","name":"Diffusers","summary":"Diffusers is Hugging Face's open-source PyTorch library for pretrained diffusion models that generate images, video, and audio. It is organized around the DiffusionPipeline API, with interchangeable noise schedulers and model components, and supports adapters such as LoRA. Version 0.40.0 (August 2026) removed JAX/Flax support.","keywords":"0.40.0 framework inference training Hugging Face Apache-2.0 Hugging Face"},{"type":"model-family","slug":"dinov3","href":"/open/dinov3/","name":"DINOv3","summary":"DINOv3 is Meta's third generation of self-supervised vision backbones, announced in August 2025. The suite includes a 6.7B-parameter ViT-7B/16, distilled Vision Transformers from 21M to 840M parameters, ConvNeXt models, and two backbones trained on satellite imagery, plus task heads for classification, depth, detection, and segmentation. The web-image models were trained on LVD-1689M, a curated set of about 1.7 billion images.","keywords":"model computer-vision self-supervised Meta (FAIR) Meta"},{"type":"model-release","slug":"dinov3-vit7b16","href":"/open/dinov3-vit7b16/","name":"DINOv3 ViT-7B/16 (LVD-1689M)","summary":"The largest DINOv3 backbone: a 6.7-billion-parameter Vision Transformer with 16-pixel patches, trained by Meta with self-supervised learning (a DINO self-distillation loss and an iBOT masked-image-modeling loss) on LVD-1689M. Training ran in three stages: pretraining, Gram anchoring, and high-resolution adaptation.","keywords":"ViT-7B/16, pretrain LVD-1689M model computer-vision self-supervised Meta (FAIR) DINOv3 License Meta DINOv3"},{"type":"software","slug":"dioptra","href":"/open/dioptra/","name":"Dioptra","summary":"Dioptra is an open-source software test platform for assessing trustworthy characteristics of AI models, built by NIST's National Cybersecurity Center of Excellence and announced by NIST in July 2024. It runs as a set of containerized services with a REST API that can be used through a web interface or a Python client to design, run, and track experiments, and its README says it supports the Measure function of the NIST AI Risk Management Framework.","keywords":"framework evaluation security National Institute of Standards and Technology (NIST) CC-BY-4.0 National Institute of Standards and Technology"},{"type":"software","slug":"docling","href":"/open/docling/","name":"Docling","summary":"Docling is an open-source Python toolkit that converts documents such as PDF, DOCX, PPTX, XLSX, HTML, images, audio, and other formats into a structured DoclingDocument that can be exported as Markdown, HTML, or JSON for generative AI applications. It includes PDF layout, reading-order, and table-structure analysis, OCR, and optional vision-language model pipelines. The project began at IBM Research in Zurich and has been hosted by the LF AI & Data Foundation since 2025.","keywords":"2.131.0 framework document-processing retrieval-augmented-generation LF AI & Data Foundation (hosted project) IBM Research (originating team) MIT LF AI & Data Foundation The Linux Foundation IBM"},{"type":"dataset","slug":"dolma","href":"/open/dolma/","name":"Dolma","summary":"Dolma is Ai2's family of English pretraining corpora built from web pages, academic publications, code, and encyclopedic text, used to train the OLMo models. The first release (v1, August 2023) was updated through v1.7 (April 2024); Dolma 3, used for Olmo 3, consists of a pool of about 9.3 trillion tokens and curated pretraining, mid-training, and long-context mixes. A Dolma 3.5 pool with additional sources and filtering has since been published.","keywords":"dataset pretraining-data Ai2 (Allen Institute for AI) ODC-By-1.0 Apache-2.0 Ai2 (Allen Institute for AI)"},{"type":"software","slug":"dspy","href":"/open/dspy/","name":"DSPy","summary":"DSPy is an open-source Python framework for building language-model programs from compositional Python modules instead of hand-written prompts. It includes optimizer algorithms that tune the prompts and weights of those programs. The project started at the Stanford NLP group.","keywords":"3.4.0 framework prompt-optimization agents retrieval-augmented-generation Stanford NLP Group MIT Stanford University"},{"type":"model-release","slug":"embeddinggemma-300m","href":"/open/embeddinggemma-300m/","name":"EmbeddingGemma 300M","summary":"EmbeddingGemma is a multilingual text embedding model in the Gemma family, built from Gemma 3 with T5Gemma initialization. Google's documentation gives its size as 308M parameters; the Hugging Face card calls it a 300M-parameter model. It accepts up to 2,048 input tokens and outputs 768-dimensional vectors, which can be truncated to 512, 256, or 128 dimensions through Matryoshka Representation Learning. Google's release page dates its release to September 4, 2025.","keywords":"embeddinggemma-300m model embedding multilingual local-inference Google DeepMind Gemma Terms of Use Google DeepMind Google (Alphabet) Gemma"},{"type":"dataset","slug":"epoch-ai-models","href":"/open/epoch-ai-models/","name":"Epoch AI Data on AI Models","summary":"A public database of machine learning models maintained by Epoch AI, recording for each model estimates such as training compute, parameter count, training dataset size, training cost, training duration, and power draw. It is published as an interactive explorer and as CSV files covering notable models, large-scale models (trained with at least 10^23 FLOP), frontier models (top five in training compute when released), and all models.","keywords":"dataset training-compute ai-trends Epoch AI CC-BY-4.0 Epoch AI"},{"type":"model-family","slug":"esm","href":"/open/esm/","name":"ESM (EvolutionaryScale and Biohub protein models)","summary":"Protein language models developed by EvolutionaryScale and, since its team joined Biohub in November 2025, released by Biohub. The line includes ESM3 (June 2024), a generative model that reasons over protein sequence, structure, and function, with a largest version of 98B parameters and a small open model; ESM C (ESM Cambrian, December 2024), representation models at 300M, 600M, and 6B parameters; and, from May 2026, ESMFold2 structure prediction models and sparse autoencoders built on ESM C.","keywords":"model language-model multimodal Biohub (Chan Zuckerberg Biohub, Inc.) EvolutionaryScale (original developer; now part of Biohub) Biohub (Chan Zuckerberg Biohub) EvolutionaryScale"},{"type":"model-release","slug":"esm3-sm-open-v1","href":"/open/esm3-sm-open-v1/","name":"ESM3 open small (esm3-sm-open-v1)","summary":"The openly released small ESM3 model, which EvolutionaryScale describes as a 1.4B-parameter open model. ESM3 is a generative masked language model that takes partial prompts across protein sequence, structure, and function tracks and predicts all three. EvolutionaryScale released it in June 2024; Biohub now distributes it under the MIT license.","keywords":"esm3-sm-open-v1 model language-model multimodal Biohub (Chan Zuckerberg Biohub, Inc.) EvolutionaryScale (original developer; now part of Biohub) MIT Biohub (Chan Zuckerberg Biohub) EvolutionaryScale ESM (EvolutionaryScale and Biohub protein models)"},{"type":"model-release","slug":"esmc-6b","href":"/open/esmc-6b/","name":"ESMC 6B (ESM Cambrian)","summary":"The largest ESM C (ESM Cambrian) protein language model: 6 billion parameters, 80 transformer layers, and 2.37e23 training FLOPs. EvolutionaryScale introduced ESM C in December 2024; Biohub's May 2026 release of ESMC, ESMFold2, and ESM Atlas included the 6B weights, and ESMFold2 is trained on top of a frozen ESMC 6B.","keywords":"ESMC-6B (esmc-6b-2024-12 on the Biohub Platform API) model language-model embedding Biohub (Chan Zuckerberg Biohub, Inc.) EvolutionaryScale (original developer; now part of Biohub) MIT Biohub (Chan Zuckerberg Biohub) EvolutionaryScale ESM (EvolutionaryScale and Biohub protein models)"},{"type":"dataset","slug":"essential-web","href":"/open/essential-web/","name":"Essential-Web v1.0","summary":"Essential-Web v1.0 is a web text dataset of about 24 trillion tokens in 23.6 billion documents, built by Essential AI from 101 Common Crawl snapshots. Every document carries metadata from a 12-category taxonomy covering subject (Free Decimal Correspondence, a Dewey Decimal-inspired scheme), page type, reasoning depth, education level, and quality, so subsets can be selected with metadata filters.","keywords":"v1.0 dataset pretraining-data Essential AI ODC-By-1.0 Essential AI"},{"type":"model-family","slug":"ether0","href":"/open/ether0/","name":"ether0","summary":"ether0 is a 24B-parameter chemistry reasoning model from FutureHouse, released in June 2025. It reasons in English and answers with molecular structures written as SMILES. It was fine-tuned and trained with reinforcement learning from Mistral AI's Mistral-Small-24B-Instruct-2501.","keywords":"model language-model reasoning science FutureHouse FutureHouse"},{"type":"model-release","slug":"ether0-24b","href":"/open/ether0-24b/","name":"ether0 (24B)","summary":"The only published ether0 checkpoint: a 24B-parameter language model that reasons in English and outputs molecules as SMILES. It is fine-tuned and reinforcement-learning trained from Mistral-Small-24B-Instruct-2501, followed by safety post-training. It was posted to Hugging Face on June 4, 2025.","keywords":"futurehouse/ether0 model language-model reasoning science FutureHouse Apache-2.0 Apache-2.0 FutureHouse ether0"},{"type":"model-family","slug":"evo-2","href":"/open/evo-2/","name":"Evo 2","summary":"Evo 2 is a family of DNA language models from Arc Institute, developed with NVIDIA and collaborators at institutions including Stanford University, UCSF, UC Berkeley, Goodfire, and the University of Washington. The models use the StripedHyena 2 architecture, work at single-nucleotide resolution with context of up to 1 million base pairs, and were trained on the OpenGenome2 dataset. Checkpoints range from 1B to 40B parameters; a 20B checkpoint derived from the 40B model was added in February 2026.","keywords":"model language-model Arc Institute Arc Institute NVIDIA"},{"type":"model-release","slug":"evo-2-20b","href":"/open/evo-2-20b/","name":"Evo 2 20B","summary":"A 20-billion-parameter Evo 2 checkpoint with 1 million base-pair context, released in February 2026. Arc created it from Evo 2 40B without additional training by removing layers that logit-lens analysis showed could be dropped with little effect on loss, while keeping the unembedding layer.","keywords":"evo2_20b model language-model Arc Institute Apache-2.0 Apache-2.0 Arc Institute NVIDIA Evo 2"},{"type":"model-release","slug":"evo-2-40b","href":"/open/evo-2-40b/","name":"Evo 2 40B","summary":"The 40-billion-parameter Evo 2 checkpoint, with 50 layers and a context length of up to 1 million base pairs. It was trained autoregressively on OpenGenome2 on 2,048 GPUs, per the Savanna README; a separate evo2_40b_base checkpoint trained at 8,192-token context is also published.","keywords":"evo2_40b model language-model Arc Institute Apache-2.0 Apache-2.0 Apache-2.0 Arc Institute NVIDIA Evo 2"},{"type":"software","slug":"executorch","href":"/open/executorch/","name":"ExecuTorch","summary":"ExecuTorch is PyTorch's open-source stack for running AI models on-device, from phones and laptops to embedded systems and microcontrollers. A model is captured with torch.export, lowered for selected hardware backends into a .pte program, and run through C++, Python, Swift/Objective-C, Kotlin/Java, or JavaScript APIs.","keywords":"runtime inference local-inference on-device Meta BSD-3-Clause Meta PyTorch Foundation"},{"type":"software","slug":"faiss","href":"/open/faiss/","name":"Faiss","summary":"Faiss is a library for similarity search and clustering of dense vectors, written in C++ with Python wrappers. It includes exact and approximate nearest-neighbor indexes, among them compressed (quantization-based) and graph-based (HNSW, NSG) indexes, and GPU implementations of several algorithms. It is developed mainly at Meta's Fundamental AI Research (FAIR) group, which released it in March 2017.","keywords":"framework vector-search numerical-computing retrieval Meta (Fundamental AI Research) MIT Meta"},{"type":"software","slug":"fastvideo","href":"/open/fastvideo/","name":"FastVideo","summary":"FastVideo is a post-training and inference framework for accelerated video generation, maintained by the Hao AI Lab at UC San Diego. Its README lists full and LoRA fine-tuning of open video diffusion transformers, Distribution Matching Distillation (DMD2), Video Sparse Attention and sparse distillation, causal distillation through Self-Forcing, and sequence-parallel distributed training and inference.","keywords":"framework video-generation training inference diffusion Hao AI Lab, UC San Diego Apache-2.0 University of California San Diego"},{"type":"model-family","slug":"fastvlm","href":"/open/fastvlm/","name":"FastVLM","summary":"FastVLM is a family of vision-language models from Apple, described in a CVPR 2025 paper, built around FastViTHD, a hybrid convolutional-transformer vision encoder that outputs fewer visual tokens to reduce encoding time for high-resolution images. Apple released 0.5B, 1.5B, and 7B variants that pair the encoder with Qwen2 language models, with PyTorch checkpoints and versions exported for Apple silicon.","keywords":"model vision-language multimodal Apple Apple"},{"type":"model-release","slug":"fastvlm-7b","href":"/open/fastvlm-7b/","name":"FastVLM 7B","summary":"FastVLM 7B is the 7B variant of Apple's FastVLM release. It combines Apple's FastViTHD vision encoder with the Qwen2-7B language model in a LLaVA-style architecture and accepts an image plus a text prompt to generate text.","keywords":"FastVLM-7B model vision-language multimodal Apple Apple Machine Learning Research Model License Agreement Apple software license (ml-fastvlm code) Apple FastVLM"},{"type":"dataset","slug":"fineweb","href":"/open/fineweb/","name":"FineWeb","summary":"FineWeb is Hugging Face's English web-text pretraining dataset, built by extracting, filtering, and deduplicating pages from Common Crawl snapshots dating back to 2013. The card describes it as more than 18.5 trillion tokens (GPT-2 tokenizer), up from about 15 trillion at first release in April 2024; version 1.4.0 (July 2025) added the Common Crawl snapshots from January to June 2025. Hugging Face also published FineWeb-Edu, an educational subset.","keywords":"v1.4.0 dataset pretraining-data Hugging Face (FineData, HuggingFaceFW) ODC-By-1.0 Apache-2.0 Hugging Face"},{"type":"model-release","slug":"florence-2-large","href":"/open/florence-2-large/","name":"Florence-2-large","summary":"Florence-2-large is the larger pretrained Florence-2 vision model, listed at 0.77B parameters on its model card. It takes an image and a task prompt and returns text, boxes, or polygons for tasks such as captioning, detection, grounding, and OCR. Since a December 2024 update, the Hugging Face repository holds a version continued-pretrained for a 4k context length on 0.1B samples, which the card says might not be trained well; the same update changed OCR output to include line separators.","keywords":"Florence-2-large model vision-language multimodal computer-vision Microsoft (Azure AI) MIT Microsoft Microsoft Florence-2"},{"type":"software","slug":"gemini-cli","href":"/open/gemini-cli/","name":"Gemini CLI","summary":"Gemini CLI is an open-source AI agent from Google that runs in a terminal and uses Google's Gemini models. It has built-in tools for file operations, shell commands, web fetching, and Google Search grounding, supports MCP servers and extensions, and can run non-interactively in scripts.","keywords":"runtime agents coding-agent developer-tools Google Apache-2.0 Google (Alphabet)"},{"type":"model-family","slug":"gemma","href":"/open/gemma/","name":"Gemma","summary":"Gemma is a family of open-weight models credited to Google DeepMind. Google's release page lists general-purpose generations including Gemma 3 (2025) and Gemma 4 (2026), along with specialized variants such as EmbeddingGemma, MedGemma, ShieldGemma, and FunctionGemma.","keywords":"model language-model multimodal Google DeepMind Google DeepMind Google (Alphabet)"},{"type":"model-release","slug":"gemma-4-12b","href":"/open/gemma-4-12b/","name":"Gemma 4 12B Unified","summary":"Gemma 4 12B Unified is an 11.95-billion-parameter Gemma 4 model with an encoder-free design: image patches and audio waveforms are projected directly into the language model instead of passing through separate encoders. It accepts text, image, audio, and video (as frames) input and generates text, with a 256K-token context window. Google's release page dates its release to June 3, 2026, after the other Gemma 4 sizes.","keywords":"4 (12B Unified) model language-model multimodal vision-language speech reasoning Google DeepMind Apache-2.0 Google DeepMind Google (Alphabet) Gemma"},{"type":"model-release","slug":"gemma-4-26b-a4b","href":"/open/gemma-4-26b-a4b/","name":"Gemma 4 26B A4B","summary":"Gemma 4 26B A4B is the mixture-of-experts model in the Gemma 4 generation, with 25.2 billion total and 3.8 billion active parameters (8 of 128 experts active, plus one shared expert). It accepts text and image input, can process video as frames, and generates text, with a 256K-token context window. Google's release page dates the initial Gemma 4 release, which included this size, to March 31, 2026.","keywords":"4 (26B A4B) model language-model multimodal vision-language reasoning mixture-of-experts Google DeepMind Apache-2.0 Google DeepMind Google (Alphabet) Gemma"},{"type":"model-release","slug":"gemma-4-31b","href":"/open/gemma-4-31b/","name":"Gemma 4 31B","summary":"Gemma 4 31B is the dense model in the Gemma 4 generation, with 30.7 billion parameters. It accepts text and image input, can process video as frames, and generates text, with a 256K-token context window. It is published as pre-trained and instruction-tuned checkpoints; Google's release page dates the initial Gemma 4 release, which included this size, to March 31, 2026.","keywords":"4 (31B) model language-model multimodal vision-language reasoning Google DeepMind Apache-2.0 Google DeepMind Google (Alphabet) Gemma"},{"type":"software","slug":"goose","href":"/open/goose/","name":"goose","summary":"goose is an open-source, general-purpose AI agent that runs on the user's machine, with a desktop app, a command-line interface, and an API, and is written in Rust. It works with multiple hosted and local model providers and connects to extensions through the Model Context Protocol. Block created goose and contributed it to the Agentic AI Foundation.","keywords":"runtime agents coding-agent automation goose, a Series of LF Projects, LLC (Agentic AI Foundation project) Apache-2.0 Agentic AI Foundation (AAIF) The Linux Foundation Block"},{"type":"software","slug":"gpt-neox","href":"/open/gpt-neox/","name":"GPT-NeoX","summary":"GPT-NeoX is EleutherAI's library for training large autoregressive language models on GPUs. It builds on NVIDIA's Megatron-LM and on DeepSpeed, and supports distributed training with ZeRO and 3D parallelism, several positional-embedding and attention options, mixture of experts, and predefined configurations for architectures such as Pythia and Llama. EleutherAI used it to train GPT-NeoX-20B and the Pythia suite.","keywords":"research-stack training distributed-computing EleutherAI Apache-2.0 EleutherAI"},{"type":"model-family","slug":"gpt-oss","href":"/open/gpt-oss/","name":"gpt-oss","summary":"gpt-oss is OpenAI's series of open-weight, text-only reasoning models, released in August 2025 in two sizes: gpt-oss-120b and gpt-oss-20b. Both are mixture-of-experts transformers trained for OpenAI's harmony response format. OpenAI later published gpt-oss-safeguard models fine-tuned from them for safety classification.","keywords":"model language-model reasoning mixture-of-experts OpenAI OpenAI"},{"type":"model-release","slug":"gpt-oss-120b","href":"/open/gpt-oss-120b/","name":"gpt-oss-120b","summary":"gpt-oss-120b is the larger gpt-oss model: a 36-layer mixture-of-experts transformer with 116.8B total and about 5.1B active parameters per token, using 128 experts with the top 4 selected per token. It is text-only, supports context up to 131,072 tokens, and ships with its MoE weights quantized to MXFP4.","keywords":"gpt-oss-120b model language-model reasoning mixture-of-experts OpenAI Apache-2.0 Apache-2.0 OpenAI gpt-oss"},{"type":"model-release","slug":"gpt-oss-20b","href":"/open/gpt-oss-20b/","name":"gpt-oss-20b","summary":"gpt-oss-20b is the smaller gpt-oss model: a 24-layer mixture-of-experts transformer with 20.9B total and about 3.6B active parameters per token, using 32 experts with the top 4 selected per token. It is text-only, supports context up to 131,072 tokens, and ships with its MoE weights quantized to MXFP4.","keywords":"gpt-oss-20b model language-model reasoning mixture-of-experts OpenAI Apache-2.0 Apache-2.0 OpenAI gpt-oss"},{"type":"model-family","slug":"gpt-oss-safeguard","href":"/open/gpt-oss-safeguard/","name":"gpt-oss-safeguard","summary":"gpt-oss-safeguard is a pair of open-weight, text-only reasoning models from OpenAI, gpt-oss-safeguard-120b and gpt-oss-safeguard-20b, post-trained from the corresponding gpt-oss models to classify content against a policy that the user supplies. OpenAI's technical report is dated October 29, 2025. OpenAI lists the models as a model partner of the Robust Open Online Safety Tools (ROOST) Model Community, a group of safety practitioners working with open models.","keywords":"model language-model reasoning mixture-of-experts safety classification OpenAI OpenAI"},{"type":"model-release","slug":"gpt-oss-safeguard-120b","href":"/open/gpt-oss-safeguard-120b/","name":"gpt-oss-safeguard-120b","summary":"gpt-oss-safeguard-120b is the larger gpt-oss-safeguard model, a text-only safety reasoning model fine-tuned from gpt-oss-120b. The model card gives 117B total parameters with 5.1B active. Its configuration lists 36 layers, 128 experts with 4 used per token, and a maximum of 131,072 position embeddings. The mixture-of-experts weights are quantized to MXFP4; attention, router, embedding, and output layers are not.","keywords":"gpt-oss-safeguard-120b model language-model reasoning mixture-of-experts safety classification OpenAI Apache-2.0 Apache-2.0 OpenAI gpt-oss-safeguard"},{"type":"model-release","slug":"gpt-oss-safeguard-20b","href":"/open/gpt-oss-safeguard-20b/","name":"gpt-oss-safeguard-20b","summary":"gpt-oss-safeguard-20b is the smaller gpt-oss-safeguard model, a text-only safety reasoning model fine-tuned from gpt-oss-20b. The model card gives 21B total parameters with 3.6B active. Its configuration lists 24 layers, 32 experts with 4 used per token, and a maximum of 131,072 position embeddings. The mixture-of-experts weights are quantized to MXFP4; attention, router, embedding, and output layers are not.","keywords":"gpt-oss-safeguard-20b model language-model reasoning mixture-of-experts safety classification OpenAI Apache-2.0 Apache-2.0 OpenAI gpt-oss-safeguard"},{"type":"software","slug":"gpt4all","href":"/open/gpt4all/","name":"GPT4All","summary":"GPT4All is Nomic's open-source desktop application for running large language models locally on Windows, macOS, and Linux. It includes LocalDocs for chatting with local files, a Python SDK, and an API server, and its Python client wraps llama.cpp.","keywords":"runtime inference local-inference Nomic AI MIT Nomic AI"},{"type":"model-release","slug":"granite-4-2-30b","href":"/open/granite-4-2-30b/","name":"Granite 4.2 30B","summary":"Granite 4.2 30B is a 30-billion-parameter dense decoder-only language model from IBM with built-in reasoning, selectable thinking, non-thinking, and low-effort modes, and tool calling. The model card states native 128K-token context support with long-context extension to 512K, while the configuration file shipped with the weights sets a maximum of 131,072 positions. Per the model card, the 30B model received a second supervised fine-tuning phase with up-sampled agentic data.","keywords":"4.2 (30B) model language-model reasoning IBM (Granite Team) Apache-2.0 Apache-2.0 CDLA-Permissive-2.0 IBM IBM Granite"},{"type":"model-release","slug":"granite-4-2-3b","href":"/open/granite-4-2-3b/","name":"Granite 4.2 3B","summary":"Granite 4.2 3B is a 3-billion-parameter dense decoder-only language model from IBM with built- in reasoning, selectable thinking, non-thinking, and low-effort modes, and tool calling. The model card states native 128K-token context support with long-context extension to 512K, while the configuration file shipped with the weights sets a maximum of 131,072 positions.","keywords":"4.2 (3B) model language-model reasoning IBM (Granite Team) Apache-2.0 Apache-2.0 CDLA-Permissive-2.0 IBM IBM Granite"},{"type":"model-release","slug":"granite-4-2-8b","href":"/open/granite-4-2-8b/","name":"Granite 4.2 8B","summary":"Granite 4.2 8B is an 8-billion-parameter dense decoder-only language model from IBM with built-in reasoning, selectable thinking, non-thinking, and low-effort modes, and tool calling. The model card states native 128K-token context support with long-context extension to 512K, while the configuration file shipped with the weights sets a maximum of 131,072 positions.","keywords":"4.2 (8B) model language-model reasoning IBM (Granite Team) Apache-2.0 Apache-2.0 CDLA-Permissive-2.0 IBM IBM Granite"},{"type":"model-family","slug":"grok","href":"/open/grok/","name":"Grok","summary":"Grok is xAI's family of large language models, first launched in November 2023 and offered mainly through the Grok assistant and the xAI API. xAI has published weights for two earlier models: Grok-1 in March 2024 and Grok 2 in August 2025. SpaceX's prospectus describes current Grok models as proprietary.","keywords":"model language-model mixture-of-experts xAI (SpaceXAI) xAI (SpaceXAI)"},{"type":"model-release","slug":"grok-2","href":"/open/grok-2/","name":"Grok 2","summary":"Grok 2 is a post-trained language model that xAI trained and used in 2024; its weights, about 500 GB in 42 files, were published on Hugging Face in August 2025. Press coverage at the time reported the release as Grok 2.5.","keywords":"Grok 2 model language-model xAI (SpaceXAI) xAI Community License Agreement xAI (SpaceXAI) Grok"},{"type":"model-release","slug":"grok-1","href":"/open/grok-1/","name":"Grok-1","summary":"Grok-1 is a 314-billion-parameter mixture-of-experts language model (8 experts, 2 used per token) that xAI trained from scratch. The published checkpoint is the raw base model from pretraining that ended in October 2023 and is not fine-tuned for dialogue or other applications.","keywords":"Grok-1 model language-model mixture-of-experts xAI (SpaceXAI) Apache-2.0 xAI (SpaceXAI) Grok"},{"type":"evaluation-tool","slug":"harvey-lab","href":"/open/harvey-lab/","name":"Harvey LAB (Legal Agent Benchmark)","summary":"LAB is an open-source benchmark from Harvey for evaluating AI agents on legal work. It consists of a task set, in which each task gives an agent instructions and a matter file of documents and defines a rubric of pass/fail criteria, and an execution harness for running agents against the tasks and grading their work product. The tasks span transactional, advisory, regulatory, and litigation work across many legal practice areas.","keywords":"eval evaluation benchmark agents legal Harvey MIT Harvey"},{"type":"evaluation-tool","slug":"helm","href":"/open/helm/","name":"HELM (Holistic Evaluation of Language Models)","summary":"HELM is an open-source Python framework from Stanford's Center for Research on Foundation Models for evaluating foundation models, including language and multimodal models. It provides benchmarks in a standardized format, one interface to models from several providers, metrics beyond accuracy such as efficiency, bias, and toxicity, and a web UI and leaderboards for inspecting results. HELM entered maintenance mode on June 1, 2026.","keywords":"0.5.16 eval evaluation benchmark Stanford Center for Research on Foundation Models (CRFM) Apache-2.0 Stanford University"},{"type":"model-family","slug":"hermes","href":"/open/hermes/","name":"Hermes","summary":"Hermes is Nous Research's series of post-trained language models, each built on a base model released by another developer. Releases listed by Nous include Nous-Hermes Llama 2 13B (July 2023), Hermes 2 models on Mistral, Mixtral, Yi, and Llama 3 bases (2023 to 2024), Hermes 3 on Llama 3.1 (August 2024), Hermes 4 at 405B, 70B, and 14B (August 2025), and Hermes 4.3 36B (December 2025). Hermes 4 and 4.3 are hybrid reasoning models that can answer directly or reason in think tags first.","keywords":"model language-model reasoning Nous Research Nous Research"},{"type":"model-release","slug":"hermes-4-70b","href":"/open/hermes-4-70b/","name":"Hermes 4 70B","summary":"A 70B-parameter hybrid reasoning model that Nous Research post-trained from Meta's Llama 3.1 70B base model, released in August 2025 with Hermes 4 405B and 14B. Nous trained it on a newly synthesized dataset of about 5 million reasoning and non-reasoning samples, and it can either answer directly or reason inside think tags first.","keywords":"Hermes-4-70B model language-model reasoning Nous Research Meta Llama 3 Community License (card license field \"llama3\") Nous Research Hermes"},{"type":"model-release","slug":"hermes-4-3-36b","href":"/open/hermes-4-3-36b/","name":"Hermes 4.3 36B","summary":"A 36B-parameter hybrid reasoning model that Nous Research post-trained from ByteDance Seed's Seed-OSS-36B-Base. It was Nous's first Hermes model trained in a decentralized way over the internet on the Psyche network, with context extended up to 512K tokens. Nous also released a centrally trained version of the same model as a research comparison.","keywords":"Hermes-4.3-36B model language-model reasoning Nous Research Apache-2.0 Nous Research Hermes"},{"type":"software","slug":"hermes-agent","href":"/open/hermes-agent/","name":"Hermes Agent","summary":"Hermes Agent is an agent runtime from Nous Research with command-line and messaging interfaces, persistent memory, reusable skills, and scheduled tasks. It can connect to different model providers and local or remote execution backends. It is separate from the Hermes family of language models.","keywords":"runtime agents automation Nous Research MIT Nous Research"},{"type":"evaluation-tool","slug":"humaneval","href":"/open/humaneval/","name":"HumanEval","summary":"HumanEval is a code-generation benchmark released by OpenAI with the Codex paper in July 2021. It contains 164 hand-written Python programming problems, each with a function signature, docstring, reference solution, and unit tests, and measures functional correctness of generated code with the pass@k metric. The GitHub repository provides the problem file and the evaluation harness.","keywords":"eval evaluation benchmark code-model OpenAI MIT OpenAI"},{"type":"evaluation-tool","slug":"humanitys-last-exam","href":"/open/humanitys-last-exam/","name":"Humanity's Last Exam (HLE)","summary":"A multimodal benchmark of 2,500 expert-written, closed-ended academic questions across more than a hundred subjects, including mathematics, the humanities, and the natural sciences, in multiple-choice and short-answer formats suited to automated grading. It was organized by teams at the Center for AI Safety and Scale AI with questions from nearly 1,000 subject-matter contributors, and was published in Nature in January 2026.","keywords":"HLE (2,500 questions, finalized April 2025); HLE-Rolling; HLE-Diamond (September 2026) eval benchmark evaluation reasoning multimodal Center for AI Safety Scale AI MIT MIT Center for AI Safety (CAIS) Scale AI"},{"type":"software","slug":"hydragnn","href":"/open/hydragnn/","name":"HydraGNN","summary":"HydraGNN is a PyTorch implementation of multi-headed graph neural networks from Oak Ridge National Laboratory, with separate output heads for graph-level and node-level properties. Its README lists distributed training with DDP, FSDP, and DeepSpeed; equivariant layers such as EGNN, PaiNN, MACE, and DimeNet; heterogeneous graph learning; global attention through GPS; and training of machine-learned interatomic potentials with energy-conserving forces.","keywords":"framework graph-neural-network training distributed-computing science Oak Ridge National Laboratory BSD-3-Clause Oak Ridge National Laboratory"},{"type":"model-family","slug":"granite","href":"/open/granite/","name":"IBM Granite","summary":"Granite is IBM's family of models. IBM lists language, speech, vision, guardian (risk detection), embedding, and time-series lines. The most recent language-model generation on IBM's Granite page is Granite 4.2: dense reasoning models in 3B, 8B, and 30B sizes, released in August 2026 and post-trained from Granite 4.1 base models.","keywords":"model language-model reasoning speech vision-language embedding IBM IBM"},{"type":"model-release","slug":"inkling-975b","href":"/open/inkling-975b/","name":"Inkling","summary":"Inkling is a 66-layer decoder-only mixture-of-experts transformer with 975B total and 41B active parameters, routing each token to 6 of 256 experts plus 2 shared experts. It accepts text, image, and audio input, outputs text, and supports a context window of up to 1M tokens. Thinking Machines Lab says it trained the model from scratch on 45 trillion tokens of text, images, audio, and video.","keywords":"Inkling model language-model multimodal mixture-of-experts reasoning Thinking Machines Lab Apache-2.0 Thinking Machines Lab Inkling"},{"type":"model-family","slug":"inkling","href":"/open/inkling/","name":"Inkling","summary":"Inkling is Thinking Machines Lab's family of open-weight multimodal mixture-of-experts models that accept text, image, and audio input and produce text. The first release, Inkling (975B total and 41B active parameters), was published on July 15, 2026, followed by Inkling-Small (276B total and 12B active) on July 30, 2026.","keywords":"model language-model multimodal mixture-of-experts reasoning Thinking Machines Lab Thinking Machines Lab"},{"type":"model-release","slug":"inkling-small","href":"/open/inkling-small/","name":"Inkling-Small","summary":"Inkling-Small is a 42-layer decoder-only mixture-of-experts transformer with 276B total and 12B active parameters, routing each token to 6 of 256 experts plus 2 shared experts. Like Inkling, it accepts text, image, and audio input, outputs text, and supports a context window of up to 1M tokens.","keywords":"Inkling-Small model language-model multimodal mixture-of-experts reasoning Thinking Machines Lab Apache-2.0 Thinking Machines Lab Inkling"},{"type":"software","slug":"hawk","href":"/open/hawk/","name":"Inspect-Hawk (Hawk)","summary":"Inspect-Hawk is METR's open-source platform for running Inspect AI evaluations on cloud infrastructure. Users define tasks, agents, and models in a YAML file; Hawk runs each evaluation in an isolated Kubernetes pod, manages model API credentials through a proxy, streams logs, stores results in a PostgreSQL warehouse, and serves a web interface for browsing them. It is built on Inspect AI, the evaluation framework created by the UK AI Security Institute.","keywords":"framework evaluation agents distributed-computing METR (Model Evaluation & Threat Research) MIT METR (Model Evaluation & Threat Research)"},{"type":"model-family","slug":"intellect","href":"/open/intellect/","name":"INTELLECT","summary":"INTELLECT is Prime Intellect's series of open-weight models. INTELLECT-1 is a 10B model trained from scratch on 1 trillion tokens with compute from distributed community contributors; INTELLECT-2 is a 32B model produced by a distributed reinforcement-learning run on QwQ-32B; INTELLECT-3 (November 2025) is a 106B mixture-of-experts model (12B active) post-trained from GLM-4.5-Air-Base with supervised fine-tuning and large-scale RL; and INTELLECT-3.1 continues INTELLECT-3 with further RL.","keywords":"model language-model reasoning mixture-of-experts Prime Intellect Prime Intellect"},{"type":"model-release","slug":"intellect-3","href":"/open/intellect-3/","name":"INTELLECT-3","summary":"A 106B-parameter mixture-of-experts reasoning model with 12B active parameters, post-trained by Prime Intellect from Z.ai's GLM-4.5-Air-Base. Training used two supervised fine-tuning stages (general reasoning, then agentic) followed by large-scale reinforcement learning on math, code, science, logic, deep-research, and software-engineering environments, run with prime-rl on a 512-GPU H200 cluster over about two months.","keywords":"INTELLECT-3 model language-model reasoning mixture-of-experts Prime Intellect MIT Apache-2.0 MIT Prime Intellect INTELLECT"},{"type":"model-release","slug":"intellect-3-1","href":"/open/intellect-3-1/","name":"INTELLECT-3.1","summary":"A 106B-parameter mixture-of-experts reasoning model (12B active) that continues the training of INTELLECT-3 with additional reinforcement learning on math, coding, software-engineering, and agentic tasks, using prime-rl and environments built with the verifiers library.","keywords":"INTELLECT-3.1 model language-model reasoning mixture-of-experts Prime Intellect MIT Apache-2.0 MIT Prime Intellect INTELLECT"},{"type":"software","slug":"jax","href":"/open/jax/","name":"JAX","summary":"JAX is a Python library for array computation and program transformation, including automatic differentiation, just-in-time compilation, and vectorization of NumPy-style code. It compiles programs with the XLA compiler to run on CPUs, GPUs, TPUs, and other accelerators.","keywords":"framework numerical-computing training compiler distributed-computing JAX core team (Google open-source project) Apache-2.0 Google (Alphabet)"},{"type":"software","slug":"kempnerforge","href":"/open/kempnerforge/","name":"KempnerForge","summary":"KempnerForge is a PyTorch-native framework from Harvard's Kempner Institute for fault-tolerant distributed training of foundation models on AI clusters. It trains decoder-only Transformers and mixture-of-experts models with FSDP2, tensor, expert, and pipeline parallelism and FP8 mixed precision, and it includes asynchronous checkpointing with auto-resume, SLURM preemption handling, activation-extraction hooks for interpretability, and vision-language model training.","keywords":"framework training distributed-computing interpretability mixture-of-experts Kempner Institute for the Study of Natural and Artificial Intelligence, Harvard University MIT Harvard University"},{"type":"software","slug":"keras","href":"/open/keras/","name":"Keras","summary":"Keras is an open-source deep learning API. Keras 3 is multi-backend: models can run on JAX, TensorFlow, or PyTorch, and an OpenVINO backend supports inference only. The keras.io site also documents KerasHub, which provides Keras 3 implementations of model architectures paired with pretrained checkpoints.","keywords":"framework training Keras team at Google Apache-2.0 Google (Alphabet)"},{"type":"model-family","slug":"laguna","href":"/open/laguna/","name":"Laguna","summary":"Laguna is Poolside's family of mixture-of-experts language models for agentic coding. Poolside's release notes and announcements list Laguna XS.2 and Laguna M.1 (April 2026), Laguna XS 2.1 (announced 2 July 2026), and Laguna S 2.1 (21 July 2026), with sizes from 33B total (3B active) to 225B total (23B active) parameters.","keywords":"model language-model code-model mixture-of-experts reasoning Poolside Poolside"},{"type":"model-release","slug":"laguna-s-2-1","href":"/open/laguna-s-2-1/","name":"Laguna S 2.1","summary":"Laguna S 2.1 is a mixture-of-experts model with 118B total and about 8B active parameters per token, built for agentic coding. It has 48 layers mixing global and sliding-window attention, 256 routed experts plus one shared expert, and a 1,048,576-token context window, with reasoning that can be switched on or off per request.","keywords":"S 2.1 model language-model code-model mixture-of-experts reasoning Poolside OpenMDW License Agreement, version 1.1 Poolside Laguna"},{"type":"model-release","slug":"laguna-xs-2-1","href":"/open/laguna-xs-2-1/","name":"Laguna XS 2.1","summary":"Laguna XS 2.1 is a mixture-of-experts model with 33B total and 3B active parameters per token, built for agentic coding on a local machine. Poolside describes it as an upgraded version of Laguna XS.2 with the same architecture. It supports interleaved reasoning between tool calls and has a 262,144-token context window.","keywords":"XS 2.1 model language-model code-model mixture-of-experts reasoning local-inference Poolside OpenMDW License Agreement, version 1.1 Poolside Laguna"},{"type":"software","slug":"langchain-framework","href":"/open/langchain-framework/","name":"LangChain (framework)","summary":"LangChain is an open-source Python framework for building agents and LLM applications. It gives a standard interface across model providers and many third-party integrations. Its agent API is built on LangGraph, and a JavaScript/TypeScript version, LangChain.js, lives in a separate repository.","keywords":"framework agents llm-applications LangChain Inc. MIT LangChain"},{"type":"software","slug":"langgraph","href":"/open/langgraph/","name":"LangGraph","summary":"LangGraph is an open-source, low-level orchestration framework and runtime for long-running, stateful agents. It is built by LangChain Inc. but can be used without the LangChain library, and a JavaScript/TypeScript version, LangGraph.js, lives in a separate repository. The README credits Pregel and Apache Beam as inspirations.","keywords":"framework agents orchestration LangChain Inc. MIT LangChain"},{"type":"software","slug":"lbann","href":"/open/lbann/","name":"LBANN (Livermore Big Artificial Neural Network Toolkit)","summary":"LBANN is an open-source deep learning training framework from Lawrence Livermore National Laboratory, built for high-performance computing systems. Its documentation describes combining model parallelism through domain decomposition with data parallelism and ensemble training, and support for supervised, self-supervised, unsupervised, and adversarial (GAN) training. The repository's main branch now holds LBANNv2, a pre-alpha Python package described as LBANN's core integration with PyTorch; the earlier toolkit remains on the v1.x branches.","keywords":"framework training distributed-computing hpc Lawrence Livermore National Laboratory (LBANN project) Apache-2.0 Lawrence Livermore National Laboratory"},{"type":"software","slug":"lemonade","href":"/open/lemonade/","name":"Lemonade","summary":"Lemonade is a local AI server that runs language, speech, and image-generation models on a PC's CPU, GPU, or NPU and exposes them through OpenAI-, Anthropic-, and Ollama-compatible APIs. It is also offered as an embeddable binary for applications, and it supports several inference engines, including llama.cpp and engines for AMD Ryzen AI NPUs.","keywords":"runtime inference local-inference serving Lemonade project (sponsored and backed by AMD, with community maintainers) Apache-2.0 AMD"},{"type":"model-family","slug":"lfm","href":"/open/lfm/","name":"LFM (Liquid Foundation Models)","summary":"LFM is Liquid AI's series of hybrid language and multimodal models for on-device use, built from gated short-convolution blocks and a small number of grouped-query attention blocks. LFM2 (from July 2025) introduced open-weight dense models from 350M parameters upward and later an 8.3B mixture-of-experts model. LFM2.5 (from January 2026) extended pretraining and scaled up reinforcement-learning post-training; 2026 releases include LFM2.5-8B-A1B (May), LFM2.5-2.6B (August), and LFM2.5-VL-3B (August), alongside vision, audio, and encoder variants.","keywords":"model language-model multimodal local-inference Liquid AI Liquid AI"},{"type":"model-release","slug":"lfm2-5-2-6b","href":"/open/lfm2-5-2-6b/","name":"LFM2.5-2.6B","summary":"A 2.69B-parameter text-only hybrid model with 30 layers (22 double-gated short-convolution blocks and 8 grouped-query attention blocks), a 131,072-token context window, and support for 16 languages. Liquid AI pretrained it on about 34 trillion tokens and post-trained it for agentic use; it always reasons before answering.","keywords":"LFM2.5-2.6B model language-model reasoning local-inference Liquid AI LFM Open License v1.0 Liquid AI LFM (Liquid Foundation Models)"},{"type":"model-release","slug":"lfm2-5-8b-a1b","href":"/open/lfm2-5-8b-a1b/","name":"LFM2.5-8B-A1B","summary":"A text-only mixture-of-experts model with 8.3B total and 1.5B active parameters, built from 18 double-gated convolution blocks and 6 grouped-query attention blocks. Liquid AI pretrained it on 38 trillion tokens, extended its context to 128,000 tokens, and tuned it for reasoning; it writes a chain of thought before its final answer. It succeeds LFM2-8B-A1B.","keywords":"LFM2.5-8B-A1B model language-model reasoning mixture-of-experts local-inference Liquid AI LFM Open License v1.0 Liquid AI LFM (Liquid Foundation Models)"},{"type":"software","slug":"liger-kernel","href":"/open/liger-kernel/","name":"Liger Kernel","summary":"Liger Kernel is an open-source collection of Triton GPU kernels for training large language models, developed at LinkedIn. It provides Hugging Face-compatible implementations of layers such as RMSNorm, RoPE, SwiGLU, and cross-entropy (including a fused linear cross-entropy), plus memory-efficient losses for post-training methods such as DPO, ORPO, and KTO.","keywords":"framework training LinkedIn BSD-2-Clause LinkedIn"},{"type":"software","slug":"litert","href":"/open/litert/","name":"LiteRT","summary":"LiteRT is Google's on-device runtime for machine learning and generative AI models on Android, iOS, desktop, web, and IoT platforms, with CPU, GPU, and NPU acceleration. LiteRT is the new name for TensorFlow Lite; it runs models converted from PyTorch, TensorFlow, and JAX.","keywords":"runtime inference local-inference on-device Google (Google AI Edge) Apache-2.0 Google (Alphabet)"},{"type":"model-family","slug":"llama","href":"/open/llama/","name":"Llama","summary":"Llama is Meta's family of large language models. Meta's Llama models repository lists generations from Llama 2 (July 2023) through Llama 4 (April 2025), each with its own license file and acceptable use policy. Llama 4 is distributed under the custom Llama 4 Community License Agreement rather than a standard open-source license.","keywords":"model language-model multimodal mixture-of-experts Meta Meta"},{"type":"model-release","slug":"llama-3-1-tulu-3-1-8b","href":"/open/llama-3-1-tulu-3-1-8b/","name":"Llama 3.1 Tülu 3.1 8B","summary":"An 8B instruction-following model from Ai2's Tülu 3 line, post-trained from Meta's Llama 3.1 8B base model through supervised fine-tuning and DPO, then reinforcement learning with verifiable rewards. Version 3.1 differs from the original Tülu 3 8B only in the final RL stage, which switched from PPO to GRPO (without a reward model) with further hyperparameter tuning.","keywords":"Llama-3.1-Tulu-3.1-8B model language-model Ai2 (Allen Institute for AI) Llama 3.1 Community License Agreement Apache-2.0 Ai2 (Allen Institute for AI) Tülu"},{"type":"model-release","slug":"llama-4-maverick-17b-128e","href":"/open/llama-4-maverick-17b-128e/","name":"Llama 4 Maverick (17Bx128E)","summary":"Llama 4 Maverick is a natively multimodal mixture-of-experts model from Meta with 17 billion active and 400 billion total parameters across 128 experts. It accepts multilingual text and images and produces text and code; the model card lists a 1M-token context length and an August 2024 knowledge cutoff. Pretrained and instruction-tuned versions were released.","keywords":"4 Maverick 17B-128E model language-model multimodal mixture-of-experts Meta Llama 4 Community License Agreement Meta Llama"},{"type":"model-release","slug":"llama-4-scout-17b-16e","href":"/open/llama-4-scout-17b-16e/","name":"Llama 4 Scout (17Bx16E)","summary":"Llama 4 Scout is a natively multimodal mixture-of-experts model from Meta with 17 billion active and 109 billion total parameters across 16 experts. It accepts multilingual text and images and produces text and code; the model card lists a 10M-token context length and an August 2024 knowledge cutoff. Pretrained and instruction-tuned versions were released.","keywords":"4 Scout 17B-16E model language-model multimodal mixture-of-experts Meta Llama 4 Community License Agreement Meta Llama"},{"type":"software","slug":"llama-cpp","href":"/open/llama-cpp/","name":"llama.cpp","summary":"llama.cpp is a C/C++ library and set of tools for running LLM and vision-language model inference locally or in the cloud, built on the ggml tensor library. It supports integer quantization and many hardware backends. It includes a command-line interface and an OpenAI-compatible server.","keywords":"runtime inference local-inference ggml team at Hugging Face (ggml.ai, acquired by Hugging Face in 2026), with the ggml-org community MIT Hugging Face"},{"type":"software","slug":"llamafile","href":"/open/llamafile/","name":"llamafile","summary":"llamafile packages an open model's weights and a llama.cpp-based inference runtime into a single executable that runs locally on several operating systems without installation. It combines llama.cpp with Cosmopolitan Libc and includes whisperfile, a single-file speech-to-text tool built on whisper.cpp.","keywords":"runtime inference local-inference Mozilla.ai Apache-2.0 Mozilla.ai"},{"type":"software","slug":"llamaindex-framework","href":"/open/llamaindex-framework/","name":"LlamaIndex (framework)","summary":"LlamaIndex is an open-source Python framework for building LLM agents, event-driven workflows, and retrieval-augmented generation (RAG) over private data. It is split into a core package and hundreds of integration packages for models, embeddings, and vector stores; the company also offers a TypeScript SDK. The README says the company now focuses mainly on its LlamaParse platform but keeps the framework available as an open toolkit.","keywords":"framework agents rag llm-applications LlamaIndex, Inc. MIT LlamaIndex"},{"type":"software","slug":"llm-d","href":"/open/llm-d/","name":"llm-d","summary":"llm-d is an open-source distributed inference serving stack for running large language models on Kubernetes. It sits above model servers such as vLLM and adds request routing that is aware of prefix caches and load, KV-cache offloading, prefill/decode disaggregation, expert parallelism for large models, autoscaling, and batch processing. Red Hat launched it in May 2025, and it became a Cloud Native Computing Foundation Sandbox project in March 2026.","keywords":"runtime inference serving distributed-computing Cloud Native Computing Foundation (Linux Foundation), Sandbox project Red Hat (launched the project; MAINTAINERS.md lists Red Hat staff in project leadership and as community managers) Google (MAINTAINERS.md lists Google staff in project leadership) IBM (IBM Research is a founding contributor; MAINTAINERS.md lists IBM staff in project leadership and as a community manager) Apache-2.0 The Linux Foundation Red Hat Google (Alphabet) IBM CoreWeave NVIDIA"},{"type":"software","slug":"llmrouter","href":"/open/llmrouter/","name":"LLMRouter","summary":"LLMRouter is an open-source Python library for routing each query to a suitable large language model, balancing response quality against inference cost. It implements more than 16 routing methods across single-round, multi-round, multimodal, agentic, and personalized routers, and adds a command-line interface, a Gradio chat interface, and a pipeline for generating routing training data. The accompanying paper introduces xRouteBench, a routing benchmark.","keywords":"framework routing inference evaluation language-model U Lab, University of Illinois Urbana-Champaign MIT University of Illinois Urbana-Champaign"},{"type":"evaluation-tool","slug":"lm-evaluation-harness","href":"/open/lm-evaluation-harness/","name":"LM Evaluation Harness","summary":"An open-source framework from EleutherAI for evaluating language models on many benchmark tasks through one interface. It supports local models (for example through Hugging Face Transformers or vLLM) and hosted model APIs, and its README lists more than 60 standard academic benchmarks with hundreds of subtasks and variants.","keywords":"0.4.13 eval evaluation benchmark EleutherAI MIT EleutherAI"},{"type":"software","slug":"marin","href":"/open/marin/","name":"Marin","summary":"Marin is an open lab and software platform for developing foundation models, covering data curation, filtering, tokenization, pretraining, post-training, and evaluation, with experiments, code, and training logs published as the work happens. It began at Stanford in 2024, originating in Stanford CRFM and HAI, and is now primarily developed by the nonprofit Open Athena with Stanford CRFM as a core collaborator. Models trained with it include Marin 8B and Marin 32B.","keywords":"research-stack training language-model pretraining-data Open Athena (Open Athena AI Foundation Inc.) Stanford Center for Research on Foundation Models (CRFM) Apache-2.0 Stanford University"},{"type":"software","slug":"markitdown","href":"/open/markitdown/","name":"MarkItDown","summary":"MarkItDown is a Microsoft Python library and command-line tool that converts files such as PDF, Word, PowerPoint, Excel, HTML, images, audio, EPUB, and ZIP archives into Markdown. It aims to keep document structure (headings, lists, tables, links) for use by language models and text analysis pipelines rather than for high-fidelity human-readable conversion.","keywords":"framework document-processing retrieval-augmented-generation Microsoft MIT Microsoft"},{"type":"software","slug":"max","href":"/open/max/","name":"MAX (Modular)","summary":"MAX is Modular's framework for serving and building AI models. It serves models through an OpenAI-compatible API, offers PyTorch-like Python APIs for writing model pipelines, and includes a GPU kernel library for NVIDIA, AMD, and Apple hardware. Its Python API, model pipelines, and GPU kernels are open source; the MAX package as a whole is distributed under the Modular Community License.","keywords":"runtime inference serving compiler Modular Apache-2.0 WITH LLVM-exception Modular Community License (MAX package usage and distribution) Modular"},{"type":"software","slug":"maxtext","href":"/open/maxtext/","name":"MaxText","summary":"MaxText is an open-source LLM library and reference implementation written in Python and JAX for training on Google Cloud TPUs and GPUs. It provides implementations of Gemma, Llama, DeepSeek, Qwen, Mistral, and other open model architectures, with support for pre-training and for post-training methods such as supervised fine-tuning and reinforcement learning.","keywords":"research-stack training distributed-computing language-model Google (AI Hypercomputer GitHub organization) Apache-2.0 Google (Alphabet)"},{"type":"model-family","slug":"medgemma","href":"/open/medgemma/","name":"MedGemma","summary":"MedGemma is a collection of Gemma 3 variants trained by Google on medical text and images, published as part of Health AI Developer Foundations (HAI-DEF). MedGemma 1 was released in May 2025 (4B multimodal and 27B text-only) with a 27B multimodal variant in July 2025, and MedGemma 1.5 4B followed in January 2026.","keywords":"model multimodal vision-language language-model medical Google (Health AI Developer Foundations) Google (Alphabet)"},{"type":"model-release","slug":"medgemma-1-5-4b-it","href":"/open/medgemma-1-5-4b-it/","name":"MedGemma 1.5 4B","summary":"MedGemma 1.5 4B is a multimodal, instruction-tuned update to MedGemma 1 4B, released by Google on January 13, 2026. It accepts text and images and produces text, adding support for CT and MRI volumes, whole-slide histopathology, longitudinal chest X-rays, anatomical localization, lab-report extraction, and text-based EHR interpretation.","keywords":"1.5.0 (4B multimodal, instruction-tuned) model multimodal vision-language medical Google (Health AI Developer Foundations) Health AI Developer Foundations Terms of Use Apache-2.0 Google (Alphabet) MedGemma"},{"type":"model-release","slug":"medgemma-27b-it","href":"/open/medgemma-27b-it/","name":"MedGemma 27B (multimodal)","summary":"MedGemma 27B multimodal is the 27B multimodal variant of MedGemma 1, released by Google on July 9, 2025. It accepts text and images and produces text, and unlike the 27B text-only variant it was also trained on medical images and FHIR-based electronic health record data. It is published only as an instruction-tuned model.","keywords":"1.0.0 (27B multimodal, instruction-tuned) model multimodal vision-language medical Google (Health AI Developer Foundations) Health AI Developer Foundations Terms of Use Apache-2.0 Google (Alphabet) MedGemma"},{"type":"software","slug":"megatron-lm","href":"/open/megatron-lm/","name":"Megatron-LM and Megatron Core","summary":"NVIDIA's Megatron-LM repository contains two components: Megatron Core, a library of GPU-optimized building blocks for training transformer models at scale (tensor, pipeline, data, expert, and context parallelism; FP16, BF16, FP8, and FP4 mixed precision), and Megatron-LM, a reference training setup with pre-configured scripts built on Megatron Core. The README says Megatron Core development moved to GitHub in December 2025, with all development and CI now happening in the open.","keywords":"research-stack training distributed-computing NVIDIA BSD-3-Clause NVIDIA"},{"type":"model-family","slug":"florence-2","href":"/open/florence-2/","name":"Microsoft Florence-2","summary":"Florence-2 is a family of vision foundation models from Microsoft's Azure AI group that handle captioning, object detection, visual grounding, segmentation, and OCR through text prompts, using a sequence-to-sequence design with a DaViT vision encoder. The models were trained on FLD-5B, a set of 5.4 billion visual annotations on 126 million images that the team built with an iterative automated annotation process. Microsoft published four checkpoints on Hugging Face in June 2024: Florence-2-base (0.23B parameters) and Florence-2-large (0.77B), plus versions of each fine-tuned on a collection of downstream tasks (base-ft and large-ft).","keywords":"model vision-language multimodal computer-vision Microsoft (Azure AI) Microsoft"},{"type":"software","slug":"mlc-llm","href":"/open/mlc-llm/","name":"MLC LLM","summary":"MLC LLM is an open-source machine learning compiler and deployment engine for large language models. Compiled models run on MLCEngine, which offers an OpenAI-compatible API through a REST server and Python, JavaScript, iOS, and Android interfaces, with GPU backends including CUDA, ROCm, Vulkan, Metal, WebGPU, and OpenCL.","keywords":"runtime inference compiler local-inference MLC team (MLC open community) CMU Catalyst (lists MLC LLM among its own research projects) Apache-2.0 Carnegie Mellon University"},{"type":"software","slug":"mlflow","href":"/open/mlflow/","name":"MLflow","summary":"MLflow is an open-source platform for managing machine learning and generative AI development. For LLM applications and agents it provides tracing and observability (with OpenTelemetry integration), evaluation, prompt management and optimization, and an AI Gateway; for model training it provides experiment tracking, model evaluation, a model registry, and deployment tools. Databricks created MLflow, which joined the Linux Foundation in June 2020.","keywords":"3.16.1 framework training evaluation llm-applications MLflow Project, a Series of LF Projects, LLC (Linux Foundation) Databricks (creator) Apache-2.0 The Linux Foundation Databricks"},{"type":"evaluation-tool","slug":"mlperf","href":"/open/mlperf/","name":"MLPerf","summary":"MLPerf is MLCommons' family of system performance benchmarks. Its suites cover training (time to train a model to a target quality), inference in datacenter, edge, mobile, and tiny settings, and also client PCs, storage, automotive, HPC training, and inference endpoints. Results are submitted in rounds, reviewed by the submitting organizations, and published together by MLCommons.","keywords":"eval benchmark evaluation training inference MLCommons Apache-2.0 Apache-2.0 Apache-2.0 MLCommons"},{"type":"software","slug":"mlx","href":"/open/mlx/","name":"MLX","summary":"MLX is an array framework for machine learning on Apple silicon from Apple machine learning research. It has a NumPy-like Python API plus C++, C, and Swift APIs, composable function transformations, lazy computation, and a unified memory model that shares arrays between the CPU and GPU.","keywords":"framework training inference numerical-computing local-inference Apple (machine learning research) MIT Apple"},{"type":"model-family","slug":"mochi","href":"/open/mochi/","name":"Mochi","summary":"Mochi is Genmo's line of text-to-video diffusion models. Its only open-weight release so far is Mochi 1 preview (October 2024), a 10B-parameter model published with its video VAE under Apache 2.0. Genmo's hosted playground refers to a Mochi 1.1 model, for which this catalog found no open-weight release.","keywords":"model multimodal Genmo Genmo"},{"type":"model-release","slug":"mochi-1-preview","href":"/open/mochi-1-preview/","name":"Mochi 1 preview","summary":"A 10B-parameter text-to-video diffusion model built on Genmo's Asymmetric Diffusion Transformer (AsymmDiT) and released as a research preview in October 2024. It is paired with AsymmVAE, a 362M-parameter video autoencoder released alongside it, and encodes prompts with a single T5-XXL model. The initial release generates 480p video.","keywords":"mochi-1-preview model multimodal Genmo Apache-2.0 Apache-2.0 Genmo Mochi"},{"type":"software","slug":"model-context-protocol","href":"/open/model-context-protocol/","name":"Model Context Protocol (MCP)","summary":"The Model Context Protocol is an open protocol for connecting AI applications to external data sources, tools, and workflows through MCP servers and clients. The project publishes the specification, a TypeScript schema with a generated JSON Schema, documentation, and official SDKs in ten programming languages. Anthropic introduced it in November 2024 and donated it to the Agentic AI Foundation in December 2025.","keywords":"framework agents protocol tool-use Model Context Protocol a Series of LF Projects, LLC (Agentic AI Foundation project) Apache-2.0 MIT CC-BY-4.0 Agentic AI Foundation (AAIF) The Linux Foundation Anthropic"},{"type":"software","slug":"mojo","href":"/open/mojo/","name":"Mojo","summary":"Mojo is a programming language developed by Modular that pairs Python-style syntax with systems programming and metaprogramming features, for writing code that runs on CPUs and GPUs. Its compiler, standard library, and tooling are published in the Mojo directory of Modular's modular repository; Mojo 1.0 was released in August 2026.","keywords":"framework compiler programming-language numerical-computing Modular Apache-2.0 WITH LLVM-exception Modular"},{"type":"model-family","slug":"molmo","href":"/open/molmo/","name":"Molmo","summary":"Molmo is Ai2's family of open vision-language models. The first release (September 2024) comprised MolmoE-1B, Molmo-7B-O, Molmo-7B-D, and Molmo-72B, trained with Ai2's PixMo data. Molmo 2 (December 2025) added video and multi-image understanding and grounding in 4B and 8B models built on Qwen3 and a 7B model (Molmo2-O-7B) built on Ai2's own Olmo 3. Ai2 has since built specialized variants on the line, including MolmoAct 2 for robot manipulation (May 2026).","keywords":"model vision-language multimodal Ai2 (Allen Institute for AI) Ai2 (Allen Institute for AI)"},{"type":"model-release","slug":"molmo2-o-7b","href":"/open/molmo2-o-7b/","name":"Molmo2-O 7B","summary":"Molmo2-O-7B is the Molmo 2 vision-language model built on Ai2's Olmo 3 7B Instruct language model with a SigLIP 2 vision encoder, released in December 2025. It accepts images, multiple images, and video with text, and can answer questions, caption, and point to or track objects. Ai2 describes it as the variant whose full model flow is open, since its language model is also Ai2's.","keywords":"Molmo2-O-7B model vision-language multimodal Ai2 (Allen Institute for AI) Apache-2.0 Apache-2.0 Ai2 (Allen Institute for AI) Molmo"},{"type":"model-family","slug":"meta-muse","href":"/open/meta-muse/","name":"Muse (Meta)","summary":"Muse is a model family from Meta Superintelligence Labs, introduced with Muse Spark in April 2026. Muse Spark versions are offered through Meta AI and the Meta Model API without public weights. Muse Glimmer, released in August 2026, is an open-weight model distilled from Muse Spark. Meta's developer site also lists Muse Image and Muse Voice Transcribe.","keywords":"model language-model multimodal reasoning Meta (Meta Superintelligence Labs) Meta"},{"type":"model-family","slug":"microsoft-muse","href":"/open/microsoft-muse/","name":"Muse (Microsoft Research WHAM)","summary":"Muse is Microsoft Research's name for its World and Human Action Model (WHAM), a generative model of gameplay trained on human play data from the Xbox game Bleeding Edge. It can generate game visuals, controller actions, or both. It was developed by Microsoft Research's Game Intelligence and Teachable AI Experiences teams with Ninja Theory.","keywords":"model multimodal Microsoft Research Microsoft"},{"type":"model-release","slug":"muse-glimmer-30b","href":"/open/muse-glimmer-30b/","name":"Muse Glimmer 30B","summary":"Muse Glimmer is an open-weight model from Meta Superintelligence Labs with about 29.6 billion parameters, including a roughly 1.8-billion-parameter perception encoder. It is a dense transformer that takes text and images as input and produces text, and was distilled from Muse Spark for agentic tasks on consumer hardware. The model card lists a context length of 131,072+ tokens and a January 4, 2026 knowledge cutoff.","keywords":"Glimmer-30B model language-model multimodal vision-language reasoning local-inference Meta (Meta Superintelligence Labs) Apache-2.0 Meta Muse (Meta)"},{"type":"dataset","slug":"nemotron-cc","href":"/open/nemotron-cc/","name":"Nemotron-CC","summary":"Nemotron-CC is NVIDIA's English pretraining dataset built from Common Crawl. The original release (December 2024) has 6.3T tokens, 4.4T globally deduplicated original tokens and 1.9T synthetic tokens, drawn from 99 Common Crawl snapshots (CC-MAIN-2013-20 through CC-MAIN-2024-30) and split into five quality buckets. Nemotron-CC-v2 (August 2025) and Nemotron-CC-v2.1 (December 2025), published on Hugging Face, add newer snapshots, further synthetic rephrasings, translated question-answer data, and other subsets.","keywords":"dataset pretraining-data NVIDIA Common Crawl Terms of Use (original Nemotron-CC release) NVIDIA Data Agreement for Model Training (Nemotron-CC-v2 and v2.1) NVIDIA"},{"type":"software","slug":"neuromancer","href":"/open/neuromancer/","name":"NeuroMANCER","summary":"NeuroMANCER (Neural Modules with Adaptive Nonlinear Constraints and Efficient Regularizations) is an open-source differentiable programming library written in PyTorch and published by Pacific Northwest National Laboratory. It is built for learning to solve parametric constrained optimization problems, physics-informed system identification, and model-based optimal control, and it provides a symbolic interface for adding physics equations, domain knowledge, and constraints to learned models.","keywords":"framework scientific-machine-learning optimization control training Pacific Northwest National Laboratory BSD-style license (Battelle Memorial Institute) Pacific Northwest National Laboratory"},{"type":"model-family","slug":"nomic-embed","href":"/open/nomic-embed/","name":"Nomic Embed","summary":"Nomic Embed is Nomic's series of open embedding models. It began with nomic-embed-text-v1 (February 2024), a text embedder with an 8,192-token context, and continued with v1.5 and its resizable Matryoshka embeddings (February 2024), vision models aligned to the v1.5 space (June 2024), the multilingual mixture-of-experts nomic-embed-text-v2-moe (February 2025), Nomic Embed Code (March 2025), and Nomic Embed Multimodal (April 2025).","keywords":"model embedding Nomic AI Nomic AI"},{"type":"model-release","slug":"nomic-embed-text-v1-5","href":"/open/nomic-embed-text-v1-5/","name":"nomic-embed-text-v1.5","summary":"A text embedding model that updates nomic-embed-text-v1 with Matryoshka representation learning, so its 768-dimension embeddings can be shortened to as few as 64 dimensions, or binarized, with a small loss in quality. Released in February 2024, it follows v1's long-context design, which Nomic describes as supporting 8,192 tokens.","keywords":"nomic-embed-text-v1.5 model embedding Nomic AI Apache-2.0 Apache-2.0 Nomic AI Nomic Embed"},{"type":"model-release","slug":"nomic-embed-text-v2-moe","href":"/open/nomic-embed-text-v2-moe/","name":"nomic-embed-text-v2-moe","summary":"A multilingual text embedding model that uses a mixture-of-experts design (8 experts, top-2 routing), with 475M total and 305M active parameters. It covers about 100 languages, produces 768-dimensional embeddings that can be truncated to 256 through Matryoshka training, and accepts inputs of up to 512 tokens. Nomic released it in February 2025.","keywords":"nomic-embed-text-v2-moe model embedding mixture-of-experts Nomic AI Apache-2.0 Apache-2.0 Nomic AI Nomic Embed"},{"type":"model-family","slug":"cosmos","href":"/open/cosmos/","name":"NVIDIA Cosmos","summary":"Cosmos is NVIDIA's family of world foundation models, datasets, and tools for physical AI such as robots, autonomous vehicles, and smart infrastructure. Earlier generations were released as separate Predict, Transfer, and Reason models (for example Cosmos-Predict2.5 and Cosmos-Reason2). Cosmos 3 merges understanding and generation of text, images, video, audio, and actions in one mixture-of-transformers model family; its Super (64B) and Nano (16B) models were released on May 31, 2026, and the Edge (4B) model followed in July 2026.","keywords":"model world-model multimodal robotics video-generation NVIDIA NVIDIA"},{"type":"model-release","slug":"cosmos3-nano","href":"/open/cosmos3-nano/","name":"NVIDIA Cosmos3-Nano","summary":"Cosmos3-Nano is the 16B-parameter model in NVIDIA's Cosmos 3 family, released on May 31, 2026. It shares the family's mixture-of-transformers design, generating text autoregressively and images, video, audio, and robot actions by diffusion, from combinations of text, image, video, audio, and action-trajectory inputs.","keywords":"Cosmos3-Nano (16B) model world-model multimodal robotics video-generation NVIDIA OpenMDW License Agreement, version 1.1 (OpenMDW-1.1) OpenMDW License Agreement, version 1.1 (Cosmos-Framework code) NVIDIA NVIDIA Cosmos"},{"type":"model-release","slug":"cosmos3-super","href":"/open/cosmos3-super/","name":"NVIDIA Cosmos3-Super","summary":"Cosmos3-Super is the 64B-parameter model in NVIDIA's Cosmos 3 family, released on May 31, 2026. It uses a mixture-of-transformers design with an autoregressive tower that generates text and a diffusion tower that generates images, video, audio, and robot actions, and it takes combinations of text, image, video, audio, and action-trajectory inputs.","keywords":"Cosmos3-Super (64B) model world-model multimodal robotics video-generation NVIDIA OpenMDW License Agreement, version 1.1 (OpenMDW-1.1) OpenMDW License Agreement, version 1.1 (Cosmos-Framework code) NVIDIA NVIDIA Cosmos"},{"type":"software","slug":"cuopt","href":"/open/cuopt/","name":"NVIDIA cuOpt","summary":"cuOpt is NVIDIA's GPU-accelerated optimization engine for linear programming, quadratic programming, and vehicle routing, with beta support for mixed integer, quadratically constrained quadratic, and second-order cone programming. Its core is written in C++ with C, Python, and server APIs. NVIDIA announced in March 2025 that it would open-source cuOpt, and an NVIDIA post in June 2025 said it was available under Apache 2.0.","keywords":"framework optimization numerical-computing NVIDIA Apache-2.0 NVIDIA"},{"type":"software","slug":"nvidia-dynamo","href":"/open/nvidia-dynamo/","name":"NVIDIA Dynamo","summary":"Dynamo is NVIDIA's open-source framework for serving generative AI models across many GPUs and nodes. It sits above inference engines such as vLLM, SGLang, and TensorRT-LLM and adds disaggregated prefill and decode, KV-cache-aware request routing, KV cache offloading, and SLA-based autoscaling behind an OpenAI-compatible frontend. It is written in Rust and Python.","keywords":"runtime inference serving distributed-computing NVIDIA Apache-2.0 NVIDIA"},{"type":"model-family","slug":"gr00t","href":"/open/gr00t/","name":"NVIDIA Isaac GR00T N","summary":"Isaac GR00T N is NVIDIA's family of open vision-language-action models for humanoid and other robots. The models take camera images, language instructions, and robot state and output continuous actions, using a vision-language backbone with a diffusion-transformer action head. The repository describes N1.7 as the latest version of GR00T N1 and lists N1.5 and N1.6 as previous releases.","keywords":"model robotics vision-language-action NVIDIA NVIDIA"},{"type":"model-release","slug":"gr00t-n1-7-3b","href":"/open/gr00t-n1-7-3b/","name":"NVIDIA Isaac GR00T N1.7 3B","summary":"GR00T N1.7 is a 3B-parameter vision-language-action model from NVIDIA that maps camera images, language instructions, and robot proprioception to continuous robot actions. It replaces the earlier Eagle backbone with Cosmos-Reason2-2B and uses a flow-matching diffusion transformer as its action head. NVIDIA first released it in early access in April 2026; the repository now describes it as a general availability release.","keywords":"N1.7 (GR00T-N1.7-3B) model robotics vision-language-action NVIDIA NVIDIA Open Model License Agreement Apache-2.0 NVIDIA NVIDIA Isaac GR00T N"},{"type":"software","slug":"isaac-lab","href":"/open/isaac-lab/","name":"NVIDIA Isaac Lab","summary":"Isaac Lab is an open-source, GPU-accelerated framework for robot learning built on NVIDIA Isaac Sim. It provides robot models, ready-to-train environments, sensor simulation, and workflows for reinforcement learning, imitation learning, and motion planning. Its technical report describes it as the successor to Isaac Gym; its development started from the Orbit framework.","keywords":"framework robotics simulation training NVIDIA BSD-3-Clause Apache-2.0 NVIDIA"},{"type":"software","slug":"nemo-framework","href":"/open/nemo-framework/","name":"NVIDIA NeMo Framework","summary":"NeMo Framework is NVIDIA's open-source Python framework for pretraining, post-training, and reinforcement learning of generative AI models, organized as separate libraries in the NVIDIA-NeMo GitHub organization, including Megatron Bridge, AutoModel, NeMo RL, Curator, Evaluator, Export-Deploy, Run, and NeMo Speech. In 2026 the original NeMo repository was split and now focuses on audio, speech, and multimodal LLMs as NeMo Speech (version 3.0), and for the framework container from 26.02 onward NVIDIA points to Megatron Bridge's documentation and release notes.","keywords":"framework training distributed-computing speech NVIDIA Apache-2.0 Apache-2.0 Apache-2.0 Apache-2.0 NVIDIA"},{"type":"model-family","slug":"nemotron","href":"/open/nemotron/","name":"NVIDIA Nemotron","summary":"Nemotron is NVIDIA's family of models for building AI agents. The current generation is organized into Nano, Lightning, Super, and Ultra tiers; NVIDIA publishes the weights on Hugging Face and, for several releases, training datasets and training recipes.","keywords":"model language-model reasoning mixture-of-experts NVIDIA NVIDIA"},{"type":"model-release","slug":"nemotron-3-super-120b-a12b","href":"/open/nemotron-3-super-120b-a12b/","name":"NVIDIA Nemotron 3 Super 120B-A12B","summary":"Nemotron 3 Super is an NVIDIA language model with 120B total and 12B active parameters, using a hybrid Mamba-2 and attention architecture with latent mixture-of-experts layers and multi-token prediction, and configurable reasoning. NVIDIA released BF16, FP8, and NVFP4 checkpoints and a separate base checkpoint on Hugging Face.","keywords":"3 Super (120B-A12B), v1.0 model language-model reasoning mixture-of-experts NVIDIA NVIDIA Nemotron Open Model License Apache-2.0 NVIDIA NVIDIA Nemotron"},{"type":"model-release","slug":"nemotron-3-ultra-550b-a55b","href":"/open/nemotron-3-ultra-550b-a55b/","name":"NVIDIA Nemotron 3 Ultra 550B-A55B","summary":"Nemotron 3 Ultra is an NVIDIA language model with 550B total and 55B active parameters, built on a hybrid Mamba-2 and attention architecture with latent mixture-of-experts layers and multi-token prediction. Reasoning can be switched on or off through the chat template. NVIDIA released BF16 and NVFP4 checkpoints and a separate base checkpoint on Hugging Face.","keywords":"3 Ultra (550B-A55B), v1.0 model language-model reasoning mixture-of-experts NVIDIA OpenMDW License Agreement, version 1.1 Apache-2.0 NVIDIA NVIDIA Nemotron"},{"type":"model-release","slug":"nemotron-3-5-lightning-30b-a3b","href":"/open/nemotron-3-5-lightning-30b-a3b/","name":"NVIDIA Nemotron 3.5 Lightning 30B-A3B","summary":"Nemotron 3.5 Lightning is an NVIDIA language model with 30B total and 3B active parameters, using a hybrid Mamba-2 and attention mixture-of-experts architecture with multi-token prediction and configurable reasoning. NVIDIA released full-precision BF16 weights, an NVFP4 checkpoint for deployment, draft models for speculative decoding, and a separate base checkpoint.","keywords":"3.5 Lightning (30B-A3B), GA model language-model reasoning mixture-of-experts NVIDIA OpenMDW License Agreement, version 1.1 Apache-2.0 NVIDIA NVIDIA Nemotron"},{"type":"model-family","slug":"parakeet","href":"/open/parakeet/","name":"NVIDIA Parakeet","summary":"Parakeet is NVIDIA's family of automatic speech recognition models built on the FastConformer encoder in the NeMo toolkit, released with CTC, RNN-T, TDT, and hybrid TDT-CTC decoders. The first checkpoints were published on Hugging Face in December 2023; the card for one of them, parakeet-rnnt-1.1b, says it was developed jointly by the NVIDIA NeMo and Suno.ai teams. Later releases include the English parakeet-tdt-0.6b-v2 (April 2025), the 25-language parakeet-tdt-0.6b-v3 (August 2025), and parakeet-unified-en-0.6b (April 2026), which handles offline and streaming transcription in one model; there are also language-specific checkpoints such as Japanese, Danish, and Vietnamese.","keywords":"model speech NVIDIA (NeMo) NVIDIA"},{"type":"model-release","slug":"parakeet-tdt-0-6b-v3","href":"/open/parakeet-tdt-0-6b-v3/","name":"NVIDIA parakeet-tdt-0.6b-v3","summary":"parakeet-tdt-0.6b-v3 is a 600-million-parameter multilingual speech recognition model with a FastConformer encoder and a TDT decoder, which predicts tokens and their durations jointly. It extends parakeet-tdt-0.6b-v2 from English to 25 European languages, detects the spoken language automatically, and outputs punctuated, capitalized text with word- and segment-level timestamps. NVIDIA released it on Hugging Face on August 14, 2025.","keywords":"parakeet-tdt-0.6b-v3 model speech NVIDIA (NeMo) CC-BY-4.0 Apache-2.0 NVIDIA NVIDIA Parakeet"},{"type":"model-release","slug":"parakeet-unified-en-0-6b","href":"/open/parakeet-unified-en-0-6b/","name":"NVIDIA parakeet-unified-en-0.6b","summary":"parakeet-unified-en-0.6b is a 600M-parameter English speech recognition model that serves both offline and streaming transcription from one checkpoint. It pairs a 24-layer FastConformer encoder, trained jointly in offline and streaming modes with a mode-consistency regularization loss, with an RNN-T decoder; streaming latency can be set from 2,080 ms down to 160 ms. Output includes punctuation and capitalization. NVIDIA released it on Hugging Face on April 7, 2026.","keywords":"parakeet-unified-en-0.6b model speech NVIDIA (NeMo) NVIDIA Open Model License Agreement Apache-2.0 NVIDIA NVIDIA Parakeet"},{"type":"software","slug":"ollama","href":"/open/ollama/","name":"Ollama","summary":"Ollama is an open-source application and runtime for downloading and running open models on macOS, Windows, and Linux. It has an API, Python and JavaScript client libraries, and integrations with coding agents and other tools. Its company, Ollama Inc., also offers cloud-hosted models.","keywords":"runtime inference local-inference Ollama Inc. MIT"},{"type":"model-family","slug":"olmo","href":"/open/olmo/","name":"OLMo","summary":"OLMo (styled Olmo in recent releases) is Ai2's series of open language models. Releases have included the first OLMo 1B and 7B models (February 2024), OLMoE (September 2024), OLMo 2 (November 2024 to May 2025), and Olmo 3 at 7B and 32B (November 2025), with Olmo 3.1 32B checkpoints added in December 2025. In March 2026 Ai2 also released Olmo Hybrid, a 7B model that mixes attention layers with linear RNN layers.","keywords":"model language-model reasoning Ai2 (Allen Institute for AI) Ai2 (Allen Institute for AI)"},{"type":"model-release","slug":"olmo-3-1025-7b","href":"/open/olmo-3-1025-7b/","name":"Olmo 3 7B (base)","summary":"The pretrained base model at the 7B scale in Ai2's Olmo 3 release, trained in three stages (pretraining, mid-training, and long-context extension) on Dolma 3 data. The model card lists 5.93 trillion training tokens, 32 layers, and a 65,536-token context length.","keywords":"Olmo-3-1025-7B model language-model Ai2 (Allen Institute for AI) Apache-2.0 Apache-2.0 Ai2 (Allen Institute for AI) OLMo"},{"type":"model-release","slug":"olmo-3-7b-instruct","href":"/open/olmo-3-7b-instruct/","name":"Olmo 3 7B Instruct","summary":"A 7B chat model in Ai2's Olmo 3 release, post-trained from the Olmo 3 7B base model with supervised fine-tuning, direct preference optimization, and reinforcement learning with verifiable rewards on the Dolci datasets. It responds without a separate reasoning trace, unlike the Olmo 3 Think models.","keywords":"Olmo-3-7B-Instruct model language-model Ai2 (Allen Institute for AI) Apache-2.0 Apache-2.0 Apache-2.0 Ai2 (Allen Institute for AI) OLMo"},{"type":"model-release","slug":"olmo-3-1-32b-think","href":"/open/olmo-3-1-32b-think/","name":"Olmo 3.1 32B Think","summary":"A 32B reasoning model from Ai2's December 2025 Olmo 3.1 update. It was produced by continuing the reinforcement learning run behind Olmo 3 32B Think for about three more weeks with additional passes over the Dolci-Think-RL data, and it writes a reasoning trace before its final answer.","keywords":"Olmo-3.1-32B-Think model language-model reasoning Ai2 (Allen Institute for AI) Apache-2.0 Apache-2.0 Apache-2.0 Ai2 (Allen Institute for AI) OLMo"},{"type":"software","slug":"onnx","href":"/open/onnx/","name":"ONNX (Open Neural Network Exchange)","summary":"ONNX is an open format for representing machine learning models, covering both deep learning and traditional ML. It defines an extensible computation graph model, built-in operators, and standard data types, with a current focus on inference. Facebook and Microsoft started the ONNX community in 2017, and LF AI announced it as a graduate-level project in November 2019.","keywords":"1.23.1 framework inference compiler ONNX project (hosted by the LF AI & Data Foundation) Apache-2.0 LF AI & Data Foundation The Linux Foundation Microsoft Meta"},{"type":"software","slug":"onnx-runtime","href":"/open/onnx-runtime/","name":"ONNX Runtime","summary":"ONNX Runtime is Microsoft's open-source, cross-platform engine for running and accelerating machine-learning models in the ONNX format. It runs models exported from frameworks such as PyTorch and TensorFlow/Keras and from classical libraries such as scikit-learn, LightGBM, and XGBoost, applying graph optimizations and hardware-specific execution providers; it also offers accelerated training for transformer models on NVIDIA GPUs.","keywords":"runtime inference local-inference Microsoft MIT Microsoft"},{"type":"software","slug":"open-webui","href":"/open/open-webui/","name":"Open WebUI","summary":"Open WebUI is a self-hosted web interface for AI models that is built to run entirely offline. It connects to Ollama and OpenAI-compatible APIs and includes retrieval over local documents, role-based user and group permissions, plugins, and voice features.","keywords":"runtime local-inference chat-interface Open WebUI, Inc. Open WebUI License (BSD 3-Clause terms plus a branding clause) BSD-3-Clause MIT"},{"type":"software","slug":"openai-agents-sdk","href":"/open/openai-agents-sdk/","name":"OpenAI Agents SDK","summary":"The OpenAI Agents SDK is an open-source framework from OpenAI for building multi-agent workflows from agents, tools, handoffs, guardrails, and sessions, with built-in tracing. OpenAI describes it as a production-ready upgrade of its earlier Swarm experiment. The Python SDK has a separate JavaScript/TypeScript counterpart.","keywords":"framework agents multi-agent OpenAI MIT MIT OpenAI"},{"type":"evaluation-tool","slug":"openai-evals","href":"/open/openai-evals/","name":"OpenAI Evals","summary":"OpenAI Evals is an open-source framework from OpenAI for evaluating language models and systems built on them, together with a registry of existing evals. Evals are defined in YAML and run from the oaieval and oaievalset command-line tools against \"completion functions\", which by default call models through the OpenAI API. The repository is not archived, but activity since mid-2024 has been limited to maintenance, and its README now points users to OpenAI's hosted Evals in the OpenAI dashboard.","keywords":"eval evaluation benchmark OpenAI MIT OpenAI"},{"type":"software","slug":"openclaw","href":"/open/openclaw/","name":"OpenClaw","summary":"OpenClaw is a self-hosted AI assistant with a gateway connecting model providers, tools, and messaging channels. It provides command-line and browser interfaces and supports both hosted and local model providers.","keywords":"runtime agents automation OpenClaw Foundation MIT"},{"type":"model-family","slug":"openelm","href":"/open/openelm/","name":"OpenELM","summary":"OpenELM is a family of small language models released by Apple in April 2024 in 270M, 450M, 1.1B, and 3B parameter sizes, each as a pretrained and an instruction-tuned model. The models use layer-wise scaling to allocate parameters across transformer layers, and Apple published them together with its CoreNet training library, pre-training configurations, training logs, and evaluation instructions.","keywords":"model language-model Apple Apple"},{"type":"model-release","slug":"openelm-3b-instruct","href":"/open/openelm-3b-instruct/","name":"OpenELM 3B Instruct","summary":"OpenELM 3B Instruct is the 3B-parameter instruction-tuned model in Apple's April 2024 OpenELM release. It was pretrained with Apple's CoreNet library on about 1.8 trillion tokens of public data and then instruction-tuned on the UltraFeedback dataset using the Hugging Face Alignment Handbook.","keywords":"OpenELM-3B-Instruct model language-model Apple Apple Machine Learning Research Model License Agreement Apple software license (CoreNet) Apple OpenELM"},{"type":"software","slug":"openhands","href":"/open/openhands/","name":"OpenHands","summary":"OpenHands is an open-source platform for building and running AI coding agents. Its main repository now holds Agent Canvas, a self-hosted browser control center. Agent Canvas runs the OpenHands agent, or other agents that support the Agent Client Protocol, on local, remote, or cloud backends. The agent itself, the Python SDK, and the Agent Server are developed in the separate Software Agent SDK repository.","keywords":"framework agents coding-agent All Hands AI (OpenHands) MIT MIT All Hands AI (OpenHands)"},{"type":"software","slug":"openpi","href":"/open/openpi/","name":"openpi","summary":"openpi is Physical Intelligence's repository of code and model checkpoints for its π0, π0-FAST, and π0.5 vision-language-action models for robot control. It provides base checkpoints pre-trained on more than 10,000 hours of robot data, fine-tuned checkpoints for the ALOHA, DROID, and LIBERO setups, and JAX and PyTorch code for inference and fine-tuning.","keywords":"research-stack robotics vision-language training inference Physical Intelligence Apache-2.0 Physical Intelligence"},{"type":"software","slug":"openvino","href":"/open/openvino/","name":"OpenVINO","summary":"OpenVINO is Intel's open-source toolkit for converting, optimizing, and running deep learning models, including language and generative models. It converts models from PyTorch, TensorFlow, ONNX, PaddlePaddle, JAX, and other frameworks and runs inference on CPUs (x86 and ARM), Intel GPUs, and Intel NPUs, with APIs in C++, Python, C, and Node.js.","keywords":"runtime inference local-inference compiler Intel Apache-2.0 Intel"},{"type":"software","slug":"openxla","href":"/open/openxla/","name":"OpenXLA","summary":"OpenXLA is an open-source project that develops machine learning compiler and infrastructure components linking frameworks such as JAX, PyTorch, and TensorFlow to hardware backends. Its website lists XLA (a compiler for GPUs, CPUs, and ML accelerators), StableHLO (a portable operation set), Shardy (an MLIR-based tensor partitioner), PJRT (a device and compiler interface), XProf (profiling tools), and Tokamax.","keywords":"framework compiler training inference distributed-computing OpenXLA project maintainers (Google-led technical leadership) Apache-2.0 Apache-2.0 Apache-2.0 Google (Alphabet)"},{"type":"model-family","slug":"palmyra","href":"/open/palmyra/","name":"Palmyra","summary":"Palmyra is Writer's family of language models. The Palmyra X models (currently X4, X5, and X6) are offered through Writer's platform and API. Writer has also published Palmyra models on Hugging Face: Palmyra-X-4.3-73B and domain models for medicine and finance under its non-commercial open model license, and the small Palmyra-mini models (September 2025) under Apache 2.0.","keywords":"model language-model Writer Writer"},{"type":"model-release","slug":"palmyra-mini","href":"/open/palmyra-mini/","name":"Palmyra-mini","summary":"A small text model from Writer, fine-tuned from Qwen2.5-1.5B, that the model card lists at 1.7 billion parameters with a 131,072-token context window. It is the non-reasoning member of the three-model Palmyra-mini family that Writer released in September 2025.","keywords":"palmyra-mini model language-model reasoning Writer Apache-2.0 Writer Palmyra"},{"type":"model-release","slug":"palmyra-mini-thinking-b","href":"/open/palmyra-mini-thinking-b/","name":"Palmyra-mini-thinking-b","summary":"A small reasoning model in Writer's Palmyra-mini family, released in September 2025. Writer built it on NVIDIA's OpenReasoning-Nemotron-1.5B and applied reinforcement-learning fine-tuning; the model card lists a 131,072-token context window.","keywords":"palmyra-mini-thinking-b model language-model reasoning Writer Apache-2.0 Writer Palmyra"},{"type":"software","slug":"paper-qa","href":"/open/paper-qa/","name":"PaperQA (paper-qa)","summary":"PaperQA is a Python package and command-line tool (pqa) from FutureHouse for retrieval-augmented question answering over PDFs, text, Office documents, and source code, focused on scientific literature. Versions 5 and later, called PaperQA2, add an agent that searches papers, gathers evidence, and writes answers with in-text citations; the package moved to calendar versioning in December 2025.","keywords":"framework agents retrieval science FutureHouse Apache-2.0 FutureHouse"},{"type":"evaluation-tool","slug":"petri","href":"/open/petri/","name":"Petri (Inspect Petri)","summary":"Petri (Parallel Exploration Tool for Risky Interactions) is an open-source tool for automated alignment audits of language models. An auditor model drives multi-turn conversations with a target model from seed instructions, simulating tools and rolling back turns, and a judge model scores the transcripts. Anthropic released it in October 2025 and handed its development to Meridian Labs, an AI evaluation nonprofit, in May 2026 with the release of Petri 3.0. The repository moved from safety-research/petri to meridianlabs-ai/inspect_petri.","keywords":"eval evaluation alignment safety agents Meridian Labs MIT Anthropic"},{"type":"model-family","slug":"phi","href":"/open/phi/","name":"Phi","summary":"Phi is Microsoft's family of small language models. Generations have included Phi-3, Phi-3.5, and Phi-4, with instruction, reasoning, multimodal, and vision-language variants published on Hugging Face and Microsoft Foundry. Recent Phi-4 releases include Phi-4-mini-flash-reasoning (July 2025) and Phi-4-reasoning-vision-15B (March 2026).","keywords":"model language-model reasoning multimodal Microsoft Microsoft"},{"type":"model-release","slug":"phi-4-mini-flash-reasoning","href":"/open/phi-4-mini-flash-reasoning/","name":"Phi-4-mini-flash-reasoning","summary":"Phi-4-mini-flash-reasoning is a 3.8-billion-parameter, English, text-only Microsoft model fine-tuned for mathematical reasoning. It uses a hybrid \"SambaY\" decoder-hybrid-decoder architecture that mixes state space model layers with attention and Differential Attention, and supports a 64K-token context.","keywords":"4-mini-flash-reasoning model language-model reasoning Microsoft MIT Microsoft Phi"},{"type":"model-release","slug":"phi-4-reasoning-vision-15b","href":"/open/phi-4-reasoning-vision-15b/","name":"Phi-4-reasoning-vision-15B","summary":"Phi-4-reasoning-vision-15B is a 15-billion-parameter multimodal reasoning model from Microsoft. It combines the Phi-4-Reasoning language model with a SigLIP-2 vision encoder, takes text and images as input, produces text, and has a 16,384-token context length. It can either reason step by step or answer directly, depending on the task.","keywords":"4-reasoning-vision-15B model vision-language multimodal reasoning Microsoft MIT MIT Microsoft Phi"},{"type":"software","slug":"prime-rl","href":"/open/prime-rl/","name":"prime-rl","summary":"prime-rl is Prime Intellect's open-source framework for large-scale, asynchronous reinforcement learning and supervised fine-tuning of language models. It separates an FSDP2-based trainer, a vLLM inference service, and an orchestrator that gathers rollouts from verifiers environments, and it supports multi-node deployment on Slurm and Kubernetes.","keywords":"research-stack training distributed-computing Prime Intellect Apache-2.0 Prime Intellect"},{"type":"model-family","slug":"profluent-e1","href":"/open/profluent-e1/","name":"Profluent-E1","summary":"Profluent-E1 is a family of protein encoder language models from Profluent, announced in November 2025 in 150M, 300M, and 600M parameter sizes. The models can take homologous sequences alongside a query sequence (retrieval augmentation) or run on a single sequence, and Profluent presents them as drop-in replacements for ESM-family encoders.","keywords":"model science embedding protein-language-model Profluent Profluent"},{"type":"model-release","slug":"e1-150m","href":"/open/e1-150m/","name":"Profluent-E1 150M (E1-150m)","summary":"The 150M-parameter member of the Profluent-E1 family of retrieval-augmented protein encoder models, released in November 2025 alongside E1-600m and E1-300m. The Hugging Face safetensors metadata counts 154,423,330 parameters stored in BF16. It can encode a single protein sequence or a query preceded by homologous sequences.","keywords":"Profluent-Bio/E1-150m model science embedding protein-language-model Profluent Profluent-E1 Clickthrough License Agreement Apache-2.0 Profluent Profluent-E1"},{"type":"model-release","slug":"e1-300m","href":"/open/e1-300m/","name":"Profluent-E1 300M (E1-300m)","summary":"The 300M-parameter member of the Profluent-E1 family of retrieval-augmented protein encoder models, released in November 2025 alongside E1-600m and E1-150m. The Hugging Face safetensors metadata counts 274,317,346 parameters stored in BF16. It can encode a single protein sequence or a query preceded by homologous sequences.","keywords":"Profluent-Bio/E1-300m model science embedding protein-language-model Profluent Profluent-E1 Clickthrough License Agreement Apache-2.0 Profluent Profluent-E1"},{"type":"model-release","slug":"e1-600m","href":"/open/e1-600m/","name":"Profluent-E1 600M (E1-600m)","summary":"The 600M-parameter member of the Profluent-E1 family of retrieval-augmented protein encoder models, released in November 2025 alongside E1-300m and E1-150m. The Hugging Face safetensors metadata counts 641,438,754 parameters stored in BF16. It can encode a single protein sequence or a query preceded by homologous sequences.","keywords":"Profluent-Bio/E1-600m model science embedding protein-language-model Profluent Profluent-E1 Clickthrough License Agreement Apache-2.0 Profluent Profluent-E1"},{"type":"model-family","slug":"pythia","href":"/open/pythia/","name":"Pythia","summary":"Pythia is EleutherAI's suite of language models built for research on how models learn during training and change with scale. The main suite has eight sizes from 70M to 12B parameters, each trained once on the Pile and once on a deduplicated Pile, on the same data in the same order, with 154 checkpoints per model. The current suite was retrained and re-released in April 2023; 14M and 31M models and additional random-seed runs were added later.","keywords":"model language-model EleutherAI EleutherAI"},{"type":"model-release","slug":"pythia-12b","href":"/open/pythia-12b/","name":"Pythia 12B","summary":"The largest model in EleutherAI's Pythia suite: an English decoder-only language model with about 11.8 billion total parameters, trained on the Pile (not deduplicated) for about 300 billion tokens. EleutherAI publishes 154 checkpoints from initialization to the end of training as Hugging Face branches.","keywords":"pythia-12b model language-model EleutherAI Apache-2.0 Apache-2.0 EleutherAI Pythia"},{"type":"software","slug":"pytorch","href":"/open/pytorch/","name":"PyTorch","summary":"PyTorch is an open-source machine learning framework for tensors and dynamic neural networks in Python, with GPU acceleration and a C++ distribution (LibTorch). Meta originally developed it and released it in 2016. Since September 2022 it has been hosted by the PyTorch Foundation under the Linux Foundation.","keywords":"framework training numerical-computing distributed-computing PyTorch Foundation BSD-3-Clause PyTorch Foundation The Linux Foundation Meta"},{"type":"software","slug":"pytorch-lightning","href":"/open/pytorch-lightning/","name":"PyTorch Lightning","summary":"PyTorch Lightning is an open-source Python framework that structures PyTorch training code and handles engineering tasks such as multi-GPU and multi-node distributed training and mixed precision. The same repository contains Lightning Fabric, a lighter layer for scaling custom PyTorch training loops, and both ship in the lightning package. It was open-sourced in 2019 and is maintained by Lightning AI.","keywords":"framework training distributed-computing Lightning AI Apache-2.0 Lightning AI"},{"type":"software","slug":"qualcomm-ai-hub-models","href":"/open/qualcomm-ai-hub-models/","name":"Qualcomm AI Hub Models","summary":"A Python package and GitHub repository of machine learning models prepared for deployment on Qualcomm devices. It provides export scripts that compile, quantize where applicable, profile, and run models through Qualcomm AI Hub Workbench, end-to-end demos for most models, and sample application code. Many of the packaged models come from other developers, such as YOLO, Llama, Mistral, Phi, and Qwen variants.","keywords":"framework inference local-inference Qualcomm Technologies, Inc. BSD-3-Clause Qualcomm"},{"type":"software","slug":"ray","href":"/open/ray/","name":"Ray","summary":"Ray is an open-source distributed computing framework for AI workloads, with a core distributed runtime and libraries for data processing, training, and serving. It was developed at UC Berkeley's RISELab and became a PyTorch Foundation-hosted project in October 2025, contributed by Anyscale.","keywords":"framework distributed-computing training serving PyTorch Foundation (hosted project) Apache-2.0 PyTorch Foundation The Linux Foundation Anyscale"},{"type":"dataset","slug":"redpajama","href":"/open/redpajama/","name":"RedPajama","summary":"RedPajama is Together AI's open pretraining dataset project. RedPajama-V1 (RedPajama-Data-1T, 2023) is a 1.2-trillion-token reproduction of the LLaMA training-data recipe drawn from Common Crawl, C4, GitHub, arXiv, Wikipedia, and Stack Exchange. The current version, RedPajama-V2 (October 2023), contains over 100 billion documents from 84 Common Crawl snapshots in English, German, French, Spanish, and Italian, with more than 40 precomputed quality signals and duplicate markers; its deduplicated, annotated portion is about 30 trillion tokens.","keywords":"V2 dataset pretraining-data Together AI (Together Computer, Inc.) Apache-2.0 Together AI"},{"type":"model-family","slug":"reka-flash","href":"/open/reka-flash/","name":"Reka Flash","summary":"Reka Flash is a line of Reka language models; this record covers its open-weight releases. Reka Flash 3 (March 2025) is a 21B-parameter reasoning model that Reka says was pretrained from scratch on publicly accessible and synthetic data, then instruction-tuned and trained with reinforcement learning; it was released as open weights under the Apache License 2.0. Reka Flash 3.1 (July 2025) updated it with further reinforcement learning aimed at coding and agentic tasks, and was released with a 3.5-bit quantized version and the Reka Quant quantization library.","keywords":"model language-model reasoning Reka Reka"},{"type":"model-release","slug":"reka-flash-3-1","href":"/open/reka-flash-3-1/","name":"Reka Flash 3.1","summary":"A 21B-parameter language model released by Reka in July 2025 as an update to Reka Flash 3. Reka post-trained it with supervised fine-tuning on synthetic and public datasets followed by large-scale reinforcement learning with verifiable rewards, with a focus on coding and on serving as a base for fine-tuning on agentic tasks. It writes a reasoning trace before its answer.","keywords":"reka-flash-3.1 model language-model reasoning code-model Reka Apache-2.0 Reka Reka Flash"},{"type":"model-family","slug":"rnj","href":"/open/rnj/","name":"Rnj","summary":"Rnj is Essential AI's line of 8B-parameter dense language models trained from scratch and aimed at code and STEM work. Rnj-1 (a base model and an instruction-tuned model) was released in December 2025, followed in 2026 by Rnj-1.5 Instruct, which extends the context window from 32K to 160K tokens.","keywords":"model language-model code-model Essential AI Essential AI"},{"type":"model-release","slug":"rnj-1-instruct","href":"/open/rnj-1-instruct/","name":"Rnj-1 Instruct","summary":"The instruction-tuned version of Essential AI's Rnj-1, an 8.3B-parameter dense model trained from scratch. The base model was pretrained on 8.4T tokens at an 8K context, extended to a 32K context in a 380B-token mid-training stage, and then given a 150B-token supervised fine-tuning stage to produce this model.","keywords":"rnj-1-instruct model language-model code-model Essential AI Apache-2.0 Essential AI Rnj"},{"type":"model-release","slug":"rnj-1-5-instruct","href":"/open/rnj-1-5-instruct/","name":"Rnj-1.5 Instruct","summary":"A long-context follow-up to Rnj-1 Instruct that extends the context window from 32K to 160K tokens. Essential AI built it from the Rnj-1 base model, switching most attention layers to block-local attention with a group of global layers in the middle, and added long-context mid-training data and more software-engineering training data.","keywords":"rnj-1.5-instruct model language-model code-model Essential AI Apache-2.0 Essential AI Rnj"},{"type":"model-release","slug":"sam-2-1-hiera-large","href":"/open/sam-2-1-hiera-large/","name":"SAM 2.1 Hiera-Large","summary":"SAM 2.1 Hiera-Large is the largest checkpoint (224.4M parameters, per the README) in Meta's SAM 2.1 suite, an improved set of SAM 2 checkpoints released in September 2024. SAM 2 is a transformer with streaming memory that segments objects in images and tracks them through video from point, box, or mask prompts.","keywords":"2.1 (sam2.1_hiera_large) model computer-vision segmentation Meta (FAIR) Apache-2.0 Meta Segment Anything (SAM)"},{"type":"model-release","slug":"sam-3-1","href":"/open/sam-3-1/","name":"SAM 3.1","summary":"SAM 3.1 is a March 2026 update of Meta's SAM 3, a model that detects, segments, and tracks objects in images and video from text phrases, image exemplars, or visual prompts such as points, boxes, and masks. The update adds Object Multiplex, which groups tracked objects into shared-memory buckets and processes them jointly instead of one at a time, together with new checkpoints and inference optimizations.","keywords":"3.1 (Object Multiplex) model computer-vision segmentation Meta (Meta Superintelligence Labs) SAM License Meta Segment Anything (SAM)"},{"type":"model-family","slug":"segment-anything","href":"/open/segment-anything/","name":"Segment Anything (SAM)","summary":"Segment Anything is Meta's family of promptable segmentation models. The original SAM (April 2023) produces object masks in images from prompts such as points or boxes; SAM 2 (July 2024, updated as SAM 2.1 in September 2024) extends this to video with streaming memory; SAM 3 (November 2025, updated as SAM 3.1 in March 2026) adds detection, segmentation, and tracking of every instance of a concept named by a short text phrase or image exemplar. Licenses differ by generation: Apache 2.0 for SAM and SAM 2, and Meta's custom SAM License for SAM 3.","keywords":"model computer-vision segmentation Meta Meta"},{"type":"software","slug":"sglang","href":"/open/sglang/","name":"SGLang","summary":"SGLang is an open-source serving framework for large language models and multimodal models, designed for low-latency, high-throughput inference on anything from a single GPU to large distributed clusters. Its runtime features include RadixAttention prefix caching, prefill-decode disaggregation, speculative decoding, continuous batching, several forms of parallelism, structured outputs, and quantization.","keywords":"0.5.20 runtime inference serving LMSYS (hosting organization) RadixArk (company described as a maintainer in a post on the LMSYS blog) Apache-2.0 LMSYS (Large Model Systems Organization)"},{"type":"software","slug":"smolagents","href":"/open/smolagents/","name":"smolagents","summary":"smolagents is an open-source Python library from Hugging Face for building agents with few abstractions. Its CodeAgent writes its actions as Python code, a ToolCallingAgent uses conventional tool calling, and agents and tools can be shared through the Hugging Face Hub.","keywords":"framework agents multi-agent Hugging Face Apache-2.0 Hugging Face"},{"type":"model-family","slug":"smollm","href":"/open/smollm/","name":"SmolLM","summary":"SmolLM is Hugging Face's series of small language models. The first SmolLM release (July 2024) came in 135M, 360M, and 1.7B sizes; SmolLM2 kept those three sizes, with the 1.7B model trained on about 11 trillion tokens (paper February 2025); and SmolLM3 (July 2025) is a 3B model released as base and instruct checkpoints, with a dual-mode reasoning instruct model. The SmolVLM vision-language models are maintained in the same repository.","keywords":"model language-model reasoning local-inference Hugging Face (Smol Models Research, HuggingFaceTB) Hugging Face"},{"type":"model-release","slug":"smollm3-3b","href":"/open/smollm3-3b/","name":"SmolLM3 3B","summary":"SmolLM3-3B is the instruct model of Hugging Face's SmolLM3 release (July 2025), a 3B-parameter decoder-only transformer pretrained on about 11 trillion tokens of web, code, math, and reasoning data in stages, then mid-trained on reasoning data and aligned with supervised fine-tuning and Anchored Preference Optimization. It can answer with or without an extended reasoning trace, supports tool calling, and natively covers six European languages.","keywords":"SmolLM3-3B model language-model reasoning local-inference Hugging Face (Smol Models Research, HuggingFaceTB) Apache-2.0 Apache-2.0 Apache-2.0 Hugging Face SmolLM"},{"type":"model-family","slug":"snowflake-arctic","href":"/open/snowflake-arctic/","name":"Snowflake Arctic","summary":"Arctic is a text-in, text-and-code-out language model pretrained from scratch by the Snowflake AI Research Team. It uses a dense-MoE hybrid transformer that combines a 10B dense model with a residual 128x3.66B mixture-of-experts MLP, for 480B total and 17B active parameters with top-2 gating. Base and instruct-tuned checkpoints were released on April 24, 2024 with weights and code under Apache-2.0.","keywords":"model language-model mixture-of-experts Snowflake AI Research Snowflake"},{"type":"model-release","slug":"snowflake-arctic-base","href":"/open/snowflake-arctic-base/","name":"Snowflake Arctic Base","summary":"Arctic Base is the pretrained checkpoint of Snowflake Arctic, a dense-MoE hybrid transformer (a 10B dense model plus a residual 128x3.66B MoE MLP) with 480B total and 17B active parameters, pretrained from scratch by the Snowflake AI Research Team. It takes text input and generates text and code.","keywords":"Base model language-model mixture-of-experts Snowflake AI Research Apache-2.0 Apache-2.0 Snowflake Snowflake Arctic"},{"type":"model-release","slug":"snowflake-arctic-instruct","href":"/open/snowflake-arctic-instruct/","name":"Snowflake Arctic Instruct","summary":"Arctic Instruct is the instruct-tuned checkpoint of Snowflake Arctic, a dense-MoE hybrid transformer (a 10B dense model plus a residual 128x3.66B MoE MLP) with 480B total and 17B active parameters, developed by the Snowflake AI Research Team. It takes text input and generates text and code.","keywords":"Instruct model language-model mixture-of-experts Snowflake AI Research Apache-2.0 Apache-2.0 Snowflake Snowflake Arctic"},{"type":"software","slug":"sparkjs","href":"/open/sparkjs/","name":"Spark (sparkjs)","summary":"Spark is an open-source JavaScript renderer for 3D Gaussian splats built on THREE.js, made by World Labs. It renders splats together with ordinary meshes, loads common splat formats such as PLY, SPZ, and SPLAT, and lets splats be edited and animated on the GPU. Spark 2.0 added a level-of-detail system for streaming large splat worlds.","keywords":"framework 3d-rendering gaussian-splatting computer-vision World Labs MIT World Labs"},{"type":"software","slug":"spec-kit","href":"/open/spec-kit/","name":"Spec Kit","summary":"Spec Kit is an open-source toolkit from GitHub that gives AI coding agents structured processes and reusable templates. Its core process, spec-driven development, turns requirements into a specification, a technical plan, and a task list before implementation; bug-fixing and idea-assessment processes are available as bundled extensions.","keywords":"framework agents coding-agent developer-tools GitHub MIT GitHub"},{"type":"model-family","slug":"starcoder2","href":"/open/starcoder2/","name":"StarCoder2","summary":"StarCoder2 is a family of code generation models from the BigCode project, released on February 28, 2024 in 3B, 7B, and 15B sizes and trained on The Stack v2. Per Hugging Face's announcement, ServiceNow trained the 3B model, Hugging Face the 7B model, and NVIDIA the 15B model. The bigcode organization on Hugging Face also hosts an instruction-tuned variant, StarCoder2-15B-Instruct-v0.1.","keywords":"model language-model code-model BigCode project (open scientific collaboration) Hugging Face (BigCode co-steward) ServiceNow (BigCode co-steward) Hugging Face ServiceNow NVIDIA"},{"type":"model-release","slug":"starcoder2-15b","href":"/open/starcoder2-15b/","name":"StarCoder2-15B","summary":"The largest StarCoder2 base model, with 15B parameters, trained on more than 4 trillion tokens covering 600+ programming languages from The Stack v2. It uses grouped-query attention, a 16,384-token context window with 4,096-token sliding-window attention, and a fill-in-the-middle training objective. NVIDIA trained it with the NeMo framework on its Eos supercomputer.","keywords":"starcoder2-15b model language-model code-model BigCode project (open scientific collaboration) Hugging Face (BigCode co-steward) ServiceNow (BigCode co-steward) BigCode OpenRAIL-M v1 License Agreement Apache-2.0 Hugging Face ServiceNow NVIDIA StarCoder2"},{"type":"model-release","slug":"starcoder2-3b","href":"/open/starcoder2-3b/","name":"StarCoder2-3B","summary":"The smallest StarCoder2 base model, with about 3B parameters, trained on more than 3 trillion tokens from The Stack v2 and other sources. Its model card lists 17 programming languages, grouped-query attention, a 16,384-token context window with 4,096-token sliding-window attention, and a fill-in-the-middle objective. ServiceNow trained it for the BigCode project.","keywords":"starcoder2-3b model language-model code-model BigCode project (open scientific collaboration) ServiceNow (BigCode co-steward; trained this model) Hugging Face (BigCode co-steward) BigCode OpenRAIL-M v1 License Agreement Apache-2.0 ServiceNow Hugging Face StarCoder2"},{"type":"software","slug":"strands-agents","href":"/open/strands-agents/","name":"Strands Agents","summary":"Strands Agents is an open-source SDK from AWS for building and running AI agents in Python and TypeScript. It runs a model-driven agent loop in the developer's own process, with tools, MCP support, multi-agent patterns, memory and sessions, and tracing. Its repository also holds Strands harness, a preassembled agent, along with a CLI and the documentation site.","keywords":"framework agents multi-agent Amazon Web Services (AWS) Apache-2.0 Amazon (AWS)"},{"type":"model-release","slug":"superapriel-15b-instruct","href":"/open/superapriel-15b-instruct/","name":"SuperApriel-15B-Instruct","summary":"A 15B-parameter instruction-tuned \"supernet\" in which each of 48 decoder layers carries four token-mixer variants: full attention, sliding-window attention, Gated DeltaNet, and Kimi Delta Attention. Choosing one mixer per layer gives deployment presets that trade output quality for decoding speed from a single checkpoint. It was derived from Apriel-1.6-15b-Thinker by distillation followed by supervised fine-tuning.","keywords":"SuperApriel-15B-Instruct model language-model reasoning ServiceNow (SLAM Labs) MIT Apache-2.0 ServiceNow Apriel"},{"type":"software","slug":"tensorflow","href":"/open/tensorflow/","name":"TensorFlow","summary":"TensorFlow is an open-source, end-to-end machine learning platform with stable Python and C++ APIs, originally developed within Google Brain. With the 2.21 release in March 2026, Google said it would limit TensorFlow work to security and bug fixes, dependency updates, and community contributions, and pointed new generative AI work to Keras 3, JAX, and PyTorch.","keywords":"framework training numerical-computing distributed-computing Google (TensorFlow team) Apache-2.0 Google (Alphabet)"},{"type":"software","slug":"tensorrt-llm","href":"/open/tensorrt-llm/","name":"TensorRT-LLM","summary":"TensorRT-LLM is NVIDIA's open-source library for running large language model inference on NVIDIA GPUs. It provides a Python LLM API, optimized kernels, Python and C++ runtimes, and an online serving command (trtllm-serve).","keywords":"runtime inference serving NVIDIA Apache-2.0 NVIDIA"},{"type":"evaluation-tool","slug":"terminal-bench","href":"/open/terminal-bench/","name":"Terminal-Bench","summary":"Terminal-Bench is a benchmark of hard tasks that AI agents must complete in command-line environments; each task has its own container environment, a human-written reference solution, and verification tests. It was first released in May 2025; Terminal-Bench 2.0 (November 2025, 89 tasks) introduced the Harbor framework for running tasks, and since July 2026 it has been maintained as a \"continuous benchmark\" with versioned releases, the latest being 4.0 (August 2026).","keywords":"eval evaluation benchmark Stanford University Laude Institute Harbor (harbor-framework project) Apache-2.0 Stanford University"},{"type":"dataset","slug":"the-pile","href":"/open/the-pile/","name":"The Pile","summary":"The Pile is an 825 GiB English text dataset for language-model pretraining, assembled by EleutherAI from 22 component datasets that include web text (Pile-CC, from Common Crawl), academic and professional sources such as PubMed Central, arXiv, and FreeLaw, books, GitHub code, and dialogue. It was described in a December 2020 paper and a 2022 datasheet and is the training corpus of EleutherAI's Pythia models. Its current availability is limited; see access conditions.","keywords":"dataset pretraining-data EleutherAI MIT EleutherAI"},{"type":"model-family","slug":"timesfm","href":"/open/timesfm/","name":"TimesFM","summary":"TimesFM (Time Series Foundation Model) is a family of pretrained time-series forecasting models developed by Google Research, first described in an ICML 2024 paper. Open checkpoints have been released as versions 1.0, 2.0, 2.5 (September 2025), and 3.0 (August 2026); version 3.0 adds native multivariate forecasting and native support for past-only and past-and-future covariates.","keywords":"model time-series forecasting Google Research Google (Alphabet)"},{"type":"model-release","slug":"timesfm-2-5-200m","href":"/open/timesfm-2-5-200m/","name":"TimesFM 2.5 200M","summary":"TimesFM 2.5 is a 200M-parameter time-series forecasting model from Google Research, released in September 2025. Compared with TimesFM 2.0 it has fewer parameters (200M, down from 500M), a context length of up to 16k points, an optional 30M quantile head for continuous quantile forecasts up to a 1k horizon, and no frequency indicator input.","keywords":"2.5 (timesfm-2.5-200m) model time-series forecasting Google Research Apache-2.0 Apache-2.0 Google (Alphabet) TimesFM"},{"type":"model-release","slug":"timesfm-3-0","href":"/open/timesfm-3-0/","name":"TimesFM 3.0","summary":"TimesFM 3.0 is a Google Research time-series forecasting model released in August 2026. Google's announcement gives it 330 million parameters and describes native multivariate forecasting, support for past-only and past-and-future covariates, and probabilistic forecasts with nine quantiles. The model card lists a 20-layer transformer architecture.","keywords":"3.0 (google/timesfm-3.0-pytorch) model time-series forecasting Google Research TimesFM Non-Commercial License v1.0 Apache-2.0 Google (Alphabet) TimesFM"},{"type":"software","slug":"tinygrad","href":"/open/tinygrad/","name":"tinygrad","summary":"tinygrad is a small end-to-end deep learning framework with a tensor library and autograd, an IR and compiler that fuse and lower kernels, a JIT, and modules for neural networks, optimizers, and datasets. Its API is similar to PyTorch's, and it supports both training and inference across several accelerator backends.","keywords":"framework training inference compiler tiny corp MIT tiny corp"},{"type":"software","slug":"transformer-explainer","href":"/open/transformer-explainer/","name":"Transformer Explainer","summary":"Transformer Explainer is an open-source interactive visualization that runs the GPT-2 (small) model, which has 124 million parameters, in the web browser. Users enter their own text and follow how the Transformer's components and operations combine to predict the next token. It was created by the Polo Club of Data Science at Georgia Tech and is described in a CHI 2026 paper.","keywords":"framework education visualization language-model Polo Club of Data Science, Georgia Tech MIT Georgia Institute of Technology"},{"type":"software","slug":"transformers","href":"/open/transformers/","name":"Transformers","summary":"Transformers is Hugging Face's open-source Python library of model definitions for text, vision, audio, video, and multimodal models, used for both inference and training. Its README presents it as a shared model-definition layer that other training frameworks and inference engines build on, and it loads pretrained checkpoints from the Hugging Face Hub.","keywords":"5.17.0 framework training inference Hugging Face Apache-2.0 Hugging Face"},{"type":"model-family","slug":"trinity","href":"/open/trinity/","name":"Trinity","summary":"Trinity is Arcee AI's family of sparse mixture-of-experts language models: Trinity Nano (6B total, about 1B active parameters), Trinity Mini (26B total, 3B active), and Trinity Large (about 398B total, 13B active). Trinity Large is published as base, pre-anneal base, preview, and reasoning-tuned Thinking checkpoints.","keywords":"model language-model mixture-of-experts reasoning Arcee AI Arcee AI"},{"type":"model-release","slug":"trinity-mini","href":"/open/trinity-mini/","name":"Trinity Mini","summary":"Trinity Mini is a 26B-parameter mixture-of-experts model with 3B active parameters, 128 experts (8 active plus 1 shared), and a 128k-token context window. It was trained on 10 trillion tokens and tuned for reasoning.","keywords":"Mini model language-model reasoning mixture-of-experts Arcee AI OpenMDW License Agreement, version 1.1 Arcee AI Trinity"},{"type":"model-release","slug":"trinity-large-thinking","href":"/open/trinity-large-thinking/","name":"Trinity-Large-Thinking","summary":"Trinity-Large-Thinking is a reasoning-tuned sparse mixture-of-experts model with about 398B total and 13B active parameters per token and a 512k-token context window. It was post-trained from Trinity-Large-Base with chain-of-thought and agentic reinforcement learning, and writes its reasoning in think blocks before answering.","keywords":"Large-Thinking model language-model reasoning mixture-of-experts Arcee AI OpenMDW License Agreement, version 1.1 Arcee AI Trinity"},{"type":"software","slug":"trl","href":"/open/trl/","name":"TRL (Transformers Reinforcement Learning)","summary":"TRL is Hugging Face's open-source library for post-training transformer language models. It provides trainer classes for methods including supervised fine-tuning (SFT), Group Relative Policy Optimization (GRPO), Direct Preference Optimization (DPO), KTO, and reward modeling, with further methods marked experimental, and it is built on the Transformers library.","keywords":"1.14.1 framework training Hugging Face Apache-2.0 Hugging Face"},{"type":"model-family","slug":"tulu","href":"/open/tulu/","name":"Tülu","summary":"Tülu is Ai2's line of instruction-following models together with the open post-training recipe used to produce them. Tülu 3 (November 2024) applied supervised fine-tuning, Direct Preference Optimization, and reinforcement learning with verifiable rewards (RLVR) to Meta's Llama 3.1 base models at 8B and 70B, followed by a 405B model in January 2025 and by Tülu 3.1 8B, which changed only the final RL stage (from PPO to GRPO).","keywords":"model language-model Ai2 (Allen Institute for AI) Ai2 (Allen Institute for AI)"},{"type":"software","slug":"unsloth-library","href":"/open/unsloth-library/","name":"Unsloth (open-source library)","summary":"Unsloth is an open-source Python library for fine-tuning and reinforcement learning of language, diffusion, text-to-speech, and embedding models, supporting LoRA, QLoRA, full fine-tuning, pretraining, GRPO, and DPO, and exporting to formats such as GGUF. The same repository contains Unsloth Studio, a web interface, and the Unsloth Desktop app.","keywords":"framework training fine-tuning local-inference Unsloth Apache-2.0 AGPL-3.0-only Unsloth"},{"type":"model-family","slug":"v-jepa-2","href":"/open/v-jepa-2/","name":"V-JEPA 2","summary":"V-JEPA 2 is Meta's family of self-supervised video encoders, pretrained on video with a masked latent-feature prediction objective. Meta released it in June 2025 with ViT-L, ViT-H, and ViT-g encoders and an action-conditioned world model (V-JEPA 2-AC) post-trained on robot video, and in March 2026 added V-JEPA 2.1, a retrained family (ViT-B to a 2B-parameter ViT-G) aimed at dense, temporally consistent features.","keywords":"model computer-vision video self-supervised robotics Meta (FAIR) Meta"},{"type":"model-release","slug":"v-jepa-2-vitg-384","href":"/open/v-jepa-2-vitg-384/","name":"V-JEPA 2 ViT-g/16 (384 px)","summary":"The 1-billion-parameter ViT-g/16 encoder from Meta's V-JEPA 2, in the version that takes 64-frame clips at 384-pixel resolution. It was pretrained with self-supervision on VideoMix22M, a mix of about 22 million video and image samples. Meta's action-conditioned V-JEPA 2-AC world model was post-trained from a ViT-g/16 V-JEPA 2 encoder.","keywords":"2 ViT-g/16, 384 resolution (vjepa2-vitg-fpc64-384) model computer-vision video self-supervised Meta (FAIR) MIT Apache-2.0 Meta V-JEPA 2"},{"type":"model-release","slug":"v-jepa-2-1-vit-gigantic-384","href":"/open/v-jepa-2-1-vit-gigantic-384/","name":"V-JEPA 2.1 ViT-G/16 (384 px)","summary":"The largest V-JEPA 2.1 model, a 2-billion-parameter ViT-G/16 video and image encoder released by Meta in March 2026. V-JEPA 2.1 changes the V-JEPA 2 recipe to learn dense, temporally consistent features, using a dense predictive loss over all tokens, self-supervision at several intermediate layers, and separate tokenizers for images and videos. Meta distilled this model into smaller ViT-B and ViT-L variants.","keywords":"2.1 ViT-G/16, 384 resolution (vjepa2_1_vit_gigantic_384) model computer-vision video self-supervised Meta (FAIR) MIT Meta V-JEPA 2"},{"type":"software","slug":"vet","href":"/open/vet/","name":"Vet (Verify Everything)","summary":"Vet is an open-source tool from Imbue that uses a language model to review code changes and the conversation history of coding agents, flagging likely bugs and places where an agent's actions diverge from what the user asked for. It runs from the terminal, as an installable agent skill, or as a GitHub Action on pull requests, and works with the user's own model provider keys.","keywords":"framework agents coding-agent developer-tools code-review Imbue AGPL-3.0-only Imbue"},{"type":"software","slug":"vllm","href":"/open/vllm/","name":"vLLM","summary":"vLLM is an open-source inference and serving engine for large language models, originally built around the PagedAttention memory-management technique. It was developed in UC Berkeley's Sky Computing Lab, and UC Berkeley contributed it to the PyTorch Foundation in 2025.","keywords":"runtime inference serving PyTorch Foundation (hosted project) vLLM core maintainers (project Technical Steering Committee) Apache-2.0 PyTorch Foundation The Linux Foundation"},{"type":"software","slug":"waymax","href":"/open/waymax/","name":"Waymax","summary":"Waymax is a multi-agent driving simulator written in JAX that builds scenarios from the Waymo Open Motion Dataset and represents road users as bounding boxes rather than raw sensor data. It includes data loading, visualization, metrics, simulated agents, and adapters to reinforcement learning interfaces such as dm-env and Brax.","keywords":"framework autonomous-driving simulation robotics Waymo Waymax License Agreement for Non-Commercial Use Waymo"},{"type":"dataset","slug":"waymo-open-dataset","href":"/open/waymo-open-dataset/","name":"Waymo Open Dataset","summary":"Autonomous driving data collected by Waymo vehicles, released for research in August 2019 and since expanded into three datasets: Perception (lidar and camera data with labels for 2,030 segments), Motion (object trajectories and 3D maps for 103,354 segments), and End-to-End Driving (360-degree camera images and routing instructions for 5,000 segments). A companion repository provides the data format definitions, evaluation metrics, and TensorFlow helper code.","keywords":"dataset autonomous-driving robotics benchmark Waymo Waymo Dataset License Agreement for Non-Commercial Use (March 2025) Apache-2.0 BSD 3-clause copyright license plus a limited patent license (wdl_limited folders) Waymo"},{"type":"software","slug":"wandb-sdk","href":"/open/wandb-sdk/","name":"Weights & Biases Python SDK (wandb)","summary":"wandb is the open-source Python library and command-line tool for Weights & Biases. Training scripts and notebooks use it to record runs with their hyperparameters and metrics on a W&B server, where results can be viewed and compared; a Public API in the same SDK queries logged data afterward.","keywords":"0.30.0 framework training experiment-tracking mlops Weights & Biases (a CoreWeave company) MIT Weights & Biases CoreWeave"},{"type":"model-release","slug":"wham-1-6b","href":"/open/wham-1-6b/","name":"WHAM 1.6B (Muse)","summary":"WHAM 1.6B is the larger of the two World and Human Action Model checkpoints Microsoft Research released for Muse. It pairs a VQ-GAN image tokenizer with a transformer trained from scratch to predict game visuals and controller actions. The same repository includes a 200M-parameter checkpoint for lighter experiments.","keywords":"WHAM 1.6B v1 model multimodal Microsoft Research Microsoft Research License Terms Microsoft Muse (Microsoft Research WHAM)"},{"type":"model-family","slug":"whisper","href":"/open/whisper/","name":"Whisper","summary":"Whisper is OpenAI's family of encoder-decoder transformer models for speech recognition, speech translation into English, and language identification. The original series of English-only and multilingual checkpoints from 39M to 1550M parameters appeared in September 2022; OpenAI followed with large-v2 (December 2022), large-v3 (November 2023), and the faster large-v3-turbo (September 2024).","keywords":"model speech OpenAI OpenAI"},{"type":"model-release","slug":"whisper-large-v3","href":"/open/whisper-large-v3/","name":"Whisper large-v3","summary":"Whisper large-v3 is a 1550M-parameter multilingual speech recognition and translation model released by OpenAI in November 2023. It keeps the architecture of the earlier large models but uses 128 Mel frequency bins instead of 80 and adds a language token for Cantonese.","keywords":"large-v3 model speech OpenAI MIT Apache-2.0 OpenAI Whisper"},{"type":"model-release","slug":"whisper-large-v3-turbo","href":"/open/whisper-large-v3-turbo/","name":"Whisper large-v3-turbo","summary":"Whisper large-v3-turbo is a multilingual speech recognition model that OpenAI derived from Whisper large-v3 by cutting the decoder from 32 layers to 4 and fine-tuning for two more epochs. It has about 0.8 billion parameters (809M in the README and Hugging Face card; the repository model card lists 798M) and is the default model in the openai-whisper package.","keywords":"large-v3-turbo model speech OpenAI MIT OpenAI Whisper"},{"type":"evaluation-tool","slug":"typesafe-ai-workflowevals","href":"/open/typesafe-ai-workflowevals/","name":"WorkflowEvals (TypeSafe AI)","summary":"WorkflowEvals is TypeSafe AI's evaluation harness for four automation workflows: invoice processing, customer service, agent-trace review, and security-incident triage. Each workflow splits a written policy into narrow yes/no, choice, and score questions for a model, then combines the answers in code into actions. Runs can compare TypeSafe's Jev with models from other providers.","keywords":"eval evaluation benchmark TypeSafe AI Apache-2.0 TypeSafe AI"},{"type":"model-family","slug":"zamba","href":"/open/zamba/","name":"Zamba","summary":"Zamba is Zyphra's series of hybrid models that combine Mamba state-space layers with shared transformer attention blocks. Releases include Zamba-7B (April 2024); Zamba2 at 1.2B, 2.7B, and 7B parameters (July to October 2024), with instruction-tuned versions; and Zamba2-VL vision-language models at the same three sizes (June 2026).","keywords":"model language-model multimodal Zyphra Zyphra"},{"type":"model-release","slug":"zamba2-7b","href":"/open/zamba2-7b/","name":"Zamba2-7B","summary":"The 7B base model in Zyphra's Zamba2 series, announced in October 2024. It interleaves Mamba2 blocks with two shared attention blocks in an alternating pattern and adds LoRA projectors to the shared MLP blocks so that each shared block can specialize by depth.","keywords":"Zamba2-7B model language-model Zyphra Apache-2.0 Apache-2.0 Zyphra Zamba"},{"type":"model-release","slug":"zamba2-vl-7b","href":"/open/zamba2-vl-7b/","name":"Zamba2-VL-7B","summary":"The largest of the three Zamba2-VL vision-language models Zyphra released in June 2026, with about 8B parameters. It pairs the Zamba2-7B hybrid state-space language model with the Qwen2.5-VL vision encoder and handles single- and multi-image understanding and grounding.","keywords":"Zamba2-VL-7B model multimodal vision-language Zyphra Apache-2.0 Apache-2.0 Zyphra Zamba"},{"type":"evaluation-tool","slug":"tau-bench","href":"/open/tau-bench/","name":"τ-bench (tau2-bench)","summary":"τ-bench is Sierra's open-source simulation framework for evaluating customer-service AI agents. In each domain an agent must follow a written policy and use tools while a simulated user takes part in the conversation, in turn-based text mode or full-duplex voice mode. The current repository, which carries the τ³-bench update, covers airline, retail, telecom, and banking-knowledge domains plus a mock domain.","keywords":"1.0.1 eval evaluation benchmark Sierra MIT Sierra"},{"type":"hub","slug":"agents","href":"/hubs/agents/","name":"AI agents and agent tooling","summary":"What separates an agent from a model, and the open protocols, SDKs, and coding agents used to build agents, including the Model Context Protocol, A2A, and AGENTS.md.","keywords":"Covers software that uses models to take actions: agent protocols and file conventions, agent SDKs and frameworks, coding agents, and benchmarks that test agents. The models that agents call are separate records with their own licenses. It does not cover hosted agent products in depth or compare agent performance. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"science","href":"/hubs/science/","name":"AI for science","summary":"Biology, chemistry, and medical models in the catalog, the science-focused organizations that publish them, and the licenses, access gates, and safety measures their documents describe.","keywords":"Covers models and tools for biology, chemistry, medicine, and scientific literature, and the organizations that publish them. It notes documented licenses, access gates, and safety measures; it does not judge scientific accuracy, clinical suitability, or biosecurity risk. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"enterprise-ai","href":"/hubs/enterprise-ai/","name":"AI in enterprise software","summary":"Enterprise software and data-platform companies in the catalog, the AI products they document, the outside models those products can call, and the open releases some of them publish under separate terms.","keywords":"Covers companies that sell enterprise software or data platforms with documented AI products, the model providers those products use, and the open models and tools they publish. It describes products from their own documentation; it does not evaluate them, compare prices, or report revenue or customers. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"evaluation","href":"/hubs/evaluation/","name":"Benchmarks and evaluation","summary":"Evaluation harnesses, benchmarks, and the organizations that run them, and what a published result can and cannot tell you about who tested what, on which version, and under which conditions.","keywords":"Covers evaluation harnesses, benchmarks, safety and system-performance tests, and the organizations that maintain them. It explains how results are produced and where they stop; it does not report scores, rank models, or say which benchmark to trust. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"chips-and-compute","href":"/hubs/chips-and-compute/","name":"Chips, cloud, and compute","summary":"The hardware and services behind AI: chip designers, accelerators, memory, networking, and AI cloud providers, and why a headquarters is not a data-center location.","keywords":"Covers organizations that design chips, memory, and networking for AI, build servers, or rent computing as a cloud or inference service, plus related open software and benchmarks. It does not report capacity, power, GPU counts, prices, or data-center sites, and a headquarters is never treated as a data-center location. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"foundation-models","href":"/hubs/foundation-models/","name":"Foundation models and the labs that build them","summary":"How model developers in the catalog make their models available, hosted, as downloadable weights, or both, why a model family is not a release, and how to read access and license terms.","keywords":"Covers organizations that train and publish general-purpose models, whether hosted, as downloadable weights, or both, and the family and release records for their models. It explains access and license terms; it does not compare model quality, report results, or give legal advice. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"open-source-foundations","href":"/hubs/open-source-foundations/","name":"Open-source foundations and stewards","summary":"How foundations such as the Linux Foundation, the PyTorch Foundation, LF AI & Data, and the Agentic AI Foundation host AI projects, and why a nonprofit's tax status is separate from the openness of what it hosts.","keywords":"Covers foundations and nonprofit organizations that host, govern, or maintain open AI software, protocols, datasets, and models, including Linux Foundation directed funds. It explains legal form and governance; it does not rate how open a foundation is, and each hosted project's license is recorded on that project's own record. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"robotics","href":"/hubs/robotics/","name":"Robotics and embodied AI","summary":"Robot makers, robot foundation models, simulators, and driving data in the catalog, with the licenses they carry and the testing and safety limits their own documents state.","keywords":"Covers organizations that build robots or autonomous vehicles, robot foundation models and world models, simulation frameworks, and robotics datasets in the catalog. It reports what sources say about licenses, testing, and safety; it does not assess whether a robot or model is safe, compare performance, or cover defense uses. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"hub","slug":"local-ai","href":"/hubs/local-ai/","name":"Running AI on your own hardware","summary":"How running AI models on your own computer or device works: local runtimes, on-device frameworks, quantization, and what to check about an open-weight model before you run it.","keywords":"Covers running open-weight models on computers and devices you control: local runtimes and desktop apps, on-device frameworks, quantization methods, and models whose publishers document local use. It does not cover hosted APIs, hardware buying advice, or speed comparisons. Featured records are examples chosen to cover the subject, not a ranking or a complete list. Profiles of people in this area are on People Behind Local AI."},{"type":"hub","slug":"data-and-datasets","href":"/hubs/data-and-datasets/","name":"Training data and datasets","summary":"Open pretraining datasets in the catalog and the organizations that steward them, how many of them build on Common Crawl, and why a dataset's license and documentation are not permission to reuse what it contains.","keywords":"Covers open pretraining and research datasets in the catalog, the organizations that collect or maintain them, and the documents that state their sources, processing, licenses, and access conditions. It does not decide whether any use of a dataset is lawful and is not legal advice. Featured records are examples chosen to cover the subject, not a ranking or a complete list."},{"type":"explainer","slug":"open-weight-vs-open-source","href":"/learn/open-weight-vs-open-source/","name":"What open weight and open source actually mean","summary":"If I can download a model, what am I allowed to do with it?","keywords":""},{"type":"explainer","slug":"how-to-read-a-model-card","href":"/learn/how-to-read-a-model-card/","name":"How to read a model card","summary":"What should I check before relying on a model?","keywords":""},{"type":"explainer","slug":"hosted-or-local","href":"/learn/hosted-or-local/","name":"Choosing hosted access or local inference","summary":"Should I use an API or run a model on my own machine?","keywords":""},{"type":"explainer","slug":"inference-hardware","href":"/learn/inference-hardware/","name":"Understanding inference hardware","summary":"What determines whether a model can run on my hardware?","keywords":""},{"type":"explainer","slug":"reading-evaluations","href":"/learn/reading-evaluations/","name":"How to read an AI evaluation","summary":"What can a benchmark result tell me about my own task?","keywords":""},{"type":"explainer","slug":"training-data-disclosures","href":"/learn/training-data-disclosures/","name":"Understanding training data disclosures","summary":"What does a provider reveal about the material used to train a model?","keywords":""},{"type":"explainer","slug":"how-the-ecosystem-fits-together","href":"/learn/how-the-ecosystem-fits-together/","name":"How the American AI ecosystem fits together","summary":"How do chips, compute providers, labs, software projects, and application companies connect?","keywords":""},{"type":"explainer","slug":"agents-and-robotics","href":"/learn/agents-and-robotics/","name":"Agents and robotics without the hype","summary":"What changes when an AI system can use tools or act in the physical world?","keywords":""},{"type":"explainer","slug":"infrastructure-and-energy-claims","href":"/learn/infrastructure-and-energy-claims/","name":"Reading AI infrastructure and energy claims","summary":"What does a data-center announcement tell us about actual operating capacity?","keywords":""},{"type":"explainer","slug":"policy-and-standards-sources","href":"/learn/policy-and-standards-sources/","name":"Finding AI policy and standards sources","summary":"Where can I find the original text behind a policy claim?","keywords":""},{"type":"glossary-term","slug":"accelerator","href":"/glossary/#accelerator","name":"Accelerator (AI chip)","summary":"A processor built to speed up particular kinds of computation, working alongside a computer's general-purpose CPU. For AI, accelerators carry out the large matrix calculations used in training and inference; GPUs are one kind, and some companies design chips specifically for neural-network workloads.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"acceptable-use-policy","href":"/glossary/#acceptable-use-policy","name":"Acceptable-use policy","summary":"A document from a model's publisher that lists uses it prohibits. A license can incorporate such a policy by reference, which makes following the policy one of the license's conditions.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"benchmark","href":"/glossary/#benchmark","name":"Benchmark","summary":"A fixed set of tasks with a scoring method, used to compare models under the same conditions. A score that a publisher reports for its own model is a reported result: it has not necessarily been reproduced by anyone else, and it depends on the prompts and settings used.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"checkpoint","href":"/glossary/#checkpoint","name":"Checkpoint","summary":"A saved copy of a model's state at some point in training. A training checkpoint holds the weights and usually what is needed to resume training, such as the optimizer state, as PyTorch's tutorial on saving and loading models describes. Publishers also use the word for any downloadable set of weights, including a quantized version prepared for deployment.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"context-window","href":"/glossary/#context-window","name":"Context window","summary":"The maximum number of tokens (small pieces of text or other input) that a model can take into account in one request, usually counting both the prompt and the generated output.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"data-center","href":"/glossary/#data-center","name":"Data center","summary":"A building or campus that houses servers, storage, and networking equipment, together with the electrical and cooling systems that keep them running. Data centers built for AI hold large clusters of accelerators connected by high-speed networks.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"evaluation-harness","href":"/glossary/#evaluation-harness","name":"Evaluation harness","summary":"Software that runs a model against one or more benchmarks in a consistent way: it formats the prompts, collects the outputs, and computes the scores, so that others can re-run the same evaluation.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"fine-tuning","href":"/glossary/#fine-tuning","name":"Fine-tuning","summary":"Continuing to train an existing model on additional data so that it behaves differently, for example to follow instructions or to handle a narrower task. The result is a new set of weights, and the base model's license may still apply to it.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"foundation","href":"/glossary/#foundation","name":"Foundation (open source)","summary":"A nonprofit organization that gives open-source projects a neutral home: it can hold their money and trademarks and provide shared services, while each project keeps its own technical governance. A large foundation can also host directed funds, programs with their own name, charter, and governing board whose money the host foundation holds and administers.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"gated-download","href":"/glossary/#gated-download","name":"Gated download","summary":"A download that requires an extra step before the files are released, such as signing in, sharing contact details, or accepting terms. On Hugging Face, a gated repository can grant access automatically once a user agrees, or it can require the publisher to approve each request by hand, as described in Hugging Face's gated models documentation.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"gguf","href":"/glossary/#gguf","name":"GGUF","summary":"A single-file format that stores a model's weights together with the metadata needed to load them, for inference with the GGML library and tools built on it, such as llama.cpp. Models are usually trained in another framework and then converted to GGUF, often with quantized weights. The format is specified in the GGUF documentation in the ggml repository.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"hosted-api","href":"/glossary/#hosted-api","name":"Hosted API","summary":"Access to a model that runs on the provider's own servers. You send inputs over the network and receive outputs, but you do not receive the weights.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"inference","href":"/glossary/#inference","name":"Inference","summary":"Running a trained model to produce outputs from inputs, as opposed to training it. Inference code is the software that loads the weights and performs that computation.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"license-scope","href":"/glossary/#license-scope","name":"License scope","summary":"The specific materials a license covers. One release can have different licenses for its weights, code, data, and documentation, and a license file in a code repository does not by itself license weights published elsewhere under other terms.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"maintainer","href":"/glossary/#maintainer","name":"Maintainer","summary":"The organization or group that currently governs a project: it reviews and accepts changes and publishes releases. The maintainer can differ from the organization that first created the project.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"mixture-of-experts","href":"/glossary/#mixture-of-experts","name":"Mixture of experts (MoE)","summary":"A model design in which some layers contain many parallel sub-networks, called experts, and a small routing network sends each token to only a few of them. Because only part of the model runs for each token, the total parameter count differs from the number of active parameters. A sparsely gated form of the design was described in Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer (Shazeer et al., 2017).","keywords":"glossary term definition"},{"type":"glossary-term","slug":"model-card","href":"/glossary/#model-card","name":"Model card","summary":"A document published with a model that describes what it is, how it was trained and evaluated, what it is intended for, and its known limitations. The format was proposed in the paper Model Cards for Model Reporting (Mitchell et al., 2018). On Hugging Face, the model card is the repository's README file, with metadata such as the license, as described in its model cards documentation.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"model-family","href":"/glossary/#model-family","name":"Model family","summary":"A named line of related models released over time, often in several sizes or generations.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"model-release","href":"/glossary/#model-release","name":"Model release","summary":"One specific published model, identified by its name, size, and version, with its own files and its own terms.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"nonprofit","href":"/glossary/#nonprofit","name":"Nonprofit","summary":"An organization that is not run to make a profit for owners or shareholders. In the United States, the IRS recognizes many nonprofits as tax-exempt: section 501(c)(3) covers organizations run for charitable, scientific, educational, and similar purposes, and section 501(c)(6) covers business leagues, such as trade associations, that promote a shared business interest. In both cases, net earnings may not benefit any private shareholder or individual.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"open-source-software","href":"/glossary/#open-source-software","name":"Open source (software)","summary":"Software distributed with its source code under a license that meets the Open Source Initiative's Open Source Definition (version 1.9). Its criteria include free redistribution, access to the source code, permission to make derived works, and no discrimination against persons, groups, or fields of endeavor. Code that is merely readable on a public website is not open source on that basis alone.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"open-source-ai","href":"/glossary/#open-source-ai","name":"Open-source AI (OSI definition)","summary":"An AI system made available under terms that grant the freedoms to use, study, modify, and share it, as set out in the Open Source Initiative's Open Source AI Definition, version 1.0. The definition requires detailed information about the training data, the complete code used to train and run the system, and the parameters, each under OSI-approved terms. It does not require every piece of training data to be downloadable.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"open-stack","href":"/glossary/#open-stack","name":"Open-stack","summary":"A USASI tier for a model release that is Open-weight and also has published inference code, training code, and a training recipe, plus at least a documented composition of its training data. It is an editorial category of rubric v0.2, not an outside standard, and it does not require the training data itself to be downloadable.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"open-weight","href":"/glossary/#open-weight","name":"Open weight","summary":"A model whose trained parameters can be downloaded by the public. The term describes availability: the license that comes with the weights may still limit commercial use, redistribution, or particular uses. Others use the term with stricter requirements; the OSI's explainer on open weights, for example, describes final parameters shared under an OSI-approved license, without the training code or data.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"parent-company","href":"/glossary/#parent-company","name":"Parent company and documented control","summary":"A parent company controls another organization, its subsidiary, either directly or through other subsidiaries. In the SEC's definitions in Rule 405, control means the power to direct an entity's management and policies, whether through owning voting securities, by contract, or otherwise, so it can rest on more than share ownership.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"power-and-energy","href":"/glossary/#power-and-energy","name":"Power and energy","summary":"Power is the rate at which electricity is supplied or used at a given moment, measured in watts (W): a kilowatt (kW) is 1,000 watts, a megawatt (MW) is 1,000 kW, and a gigawatt (GW) is 1,000 MW. Energy is power used over time, measured in watthours (Wh): a kilowatthour (kWh) is one kilowatt for one hour, and a megawatthour (MWh) is one megawatt for one hour. A figure in megawatts or gigawatts therefore describes capacity at a moment, not an amount of electricity used. The U.S. Energy Information Administration explains these units in Measuring electricity.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"quantization","href":"/glossary/#quantization","name":"Quantization","summary":"Storing a model's weights, and sometimes its intermediate values, at lower numerical precision (for example 4-bit integers instead of 16-bit numbers) to reduce memory use and speed up inference, sometimes at some cost in output quality.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"restricted-weights","href":"/glossary/#restricted-weights","name":"Restricted weights","summary":"A USASI tier for a model release whose weights can be obtained only by request, with approval, or by some users, for example through a gated download that the publisher reviews. Such a release is not counted as Open-weight, however much else about it is published.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"reviewed-date","href":"/glossary/#reviewed-date","name":"Reviewed date","summary":"The date an editor last checked a piece of evidence against its sources. It is different from the date a record was edited, the date an artifact was released, and the date the site was built.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"self-hosting","href":"/glossary/#self-hosting","name":"Self-hosting / local inference","summary":"Running a model on hardware you control, such as a laptop, a workstation, or your own servers, instead of calling a provider's hosted service. It requires access to the weights and to software that can run them, and the weights' license still applies.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"subsidiary","href":"/glossary/#subsidiary","name":"Subsidiary","summary":"A company that is owned or controlled by another company, its parent. A research unit is a similar relationship within a single company.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"tokenizer","href":"/glossary/#tokenizer","name":"Tokenizer","summary":"The component that splits input text into tokens, such as words, parts of words, or single characters, and converts them into the numeric IDs a model works with. A model is normally run with the same tokenizer it was trained with. Hugging Face's guide to tokenization algorithms describes the common subword methods.","keywords":"glossary term definition"},{"type":"glossary-term","slug":"unknown","href":"/glossary/#unknown","name":"Unknown","summary":"A status meaning that an item has not been assessed yet, or that the available evidence is not enough to decide. It never stands in for \"no.\"","keywords":"glossary term definition"},{"type":"glossary-term","slug":"weights","href":"/glossary/#weights","name":"Weights","summary":"The numerical parameters a model learns during training. Together with the model's architecture and inference code they make up the model, a breakdown the Open Source AI Definition also uses. Weights are usually distributed as large files, sometimes with checkpoints saved at intermediate points in training.","keywords":"glossary term definition"},{"type":"person","slug":"0xsero","href":"/local/#0xsero","name":"0xSero","summary":"Developer of the Local AI Registry, tested recipes for running open models on specific GPUs","keywords":"Sybil Solutions"},{"type":"person","slug":"andrej-karpathy","href":"/local/#andrej-karpathy","name":"Andrej Karpathy","summary":"Wrote llama2.c, a dependency-free C program for running Llama 2 models on a personal computer","keywords":"OpenAI Tesla"},{"type":"person","slug":"andriy-mulyar","href":"/local/#andriy-mulyar","name":"Andriy Mulyar","summary":"CEO of Nomic and co-author of the GPT4All and Nomic Embed technical reports","keywords":"Nomic AI"},{"type":"person","slug":"angelos-katharopoulos","href":"/local/#angelos-katharopoulos","name":"Angelos Katharopoulos","summary":"One of the four initial developers of MLX, Apple's array framework for machine learning on Apple silicon","keywords":"Apple"},{"type":"person","slug":"awni-hannun","href":"/local/#awni-hannun","name":"Awni Hannun","summary":"Co-creator of MLX, an array framework for machine learning on Apple silicon","keywords":"Anthropic Apple"},{"type":"person","slug":"brandon-duderstadt","href":"/local/#brandon-duderstadt","name":"Brandon Duderstadt","summary":"Nomic co-founder and co-author of the GPT4All and Nomic Embed technical reports","keywords":"Calcifer Computing Nomic AI"},{"type":"person","slug":"byron-hsu","href":"/local/#byron-hsu","name":"Byron Hsu","summary":"Led Liger Kernel, open-source Triton kernels for memory-efficient LLM training and fine-tuning","keywords":"SGLang project LinkedIn"},{"type":"person","slug":"chris-lattner","href":"/local/#chris-lattner","name":"Chris Lattner","summary":"Co-founder of Modular, maker of the MAX inference framework and the Mojo language","keywords":"Modular"},{"type":"person","slug":"clement-delangue","href":"/local/#clement-delangue","name":"Clément Delangue","summary":"Co-founder and CEO of Hugging Face, whose Hub distributes open models and the libraries used to run them","keywords":"Hugging Face"},{"type":"person","slug":"daniel-han","href":"/local/#daniel-han","name":"Daniel Han","summary":"Works on Unsloth, open-source software for fine-tuning and running models on local hardware","keywords":"Unsloth"},{"type":"person","slug":"david-heinemeier-hansson","href":"/local/#david-heinemeier-hansson","name":"David Heinemeier Hansson","summary":"Creator of Omarchy, a Linux system that sets up local-model tools such as Ollama and LM Studio","keywords":"37signals Omacom Foundation"},{"type":"person","slug":"francois-chollet","href":"/local/#francois-chollet","name":"François Chollet","summary":"Creator and project lead of Keras, whose KerasHub runs open-weight models on JAX, PyTorch, or TensorFlow","keywords":"ARC Prize Foundation Ndea"},{"type":"person","slug":"georgi-gerganov","href":"/local/#georgi-gerganov","name":"Georgi Gerganov","summary":"Develops llama.cpp and the ggml tensor library for running models on local hardware","keywords":"Hugging Face"},{"type":"person","slug":"jeffrey-morgan","href":"/local/#jeffrey-morgan","name":"Jeffrey Morgan","summary":"Works on Ollama, an open-source tool for running open models on your own machine","keywords":"Ollama"},{"type":"person","slug":"jeremy-howard","href":"/local/#jeremy-howard","name":"Jeremy Howard","summary":"Co-founder of Answer.AI, which released FSDP/QLoRA for fine-tuning large models on gaming GPUs","keywords":"Answer.AI fast.ai"},{"type":"person","slug":"joshua-lochner","href":"/local/#joshua-lochner","name":"Joshua Lochner","summary":"Works on Transformers.js at Hugging Face, a library for running Transformers models in the browser","keywords":"Hugging Face"},{"type":"person","slug":"julien-chaumond","href":"/local/#julien-chaumond","name":"Julien Chaumond","summary":"CTO of Hugging Face; added the Hub's local-apps registry and works on the hf command-line tool","keywords":"Hugging Face"},{"type":"person","slug":"loubna-ben-allal","href":"/local/#loubna-ben-allal","name":"Loubna Ben Allal","summary":"Research Engineer at Hugging Face leading the SmolLM series of small open language models","keywords":"Hugging Face"},{"type":"person","slug":"luca-soldaini","href":"/local/#luca-soldaini","name":"Luca Soldaini","summary":"Former co-lead of Ai2's Olmo open-model project; now a member of the technical staff at Microsoft AI","keywords":"Microsoft AI Ai2 (Allen Institute for AI)"},{"type":"person","slug":"maxime-labonne","href":"/local/#maxime-labonne","name":"Maxime Labonne","summary":"Head of Post-Training at Liquid AI; author of the LLM Course and its quantization tutorials","keywords":"Liquid AI"},{"type":"person","slug":"mia","href":"/local/#mia","name":"Mia","summary":"Developer of local model deployment recipes and the sparkDash monitoring dashboard","keywords":"Mia's AI Lab"},{"type":"person","slug":"michael-han","href":"/local/#michael-han","name":"Michael Han","summary":"Works on Unsloth and is an area owner for the Studio UI, its web interface for running and training models","keywords":"Unsloth"},{"type":"person","slug":"omar-sanseviero","href":"/local/#omar-sanseviero","name":"Omar Sanseviero","summary":"Developer Experience Lead at Google DeepMind, working on Gemma launches and developer guides","keywords":"Google DeepMind Hugging Face"},{"type":"person","slug":"soumith-chintala","href":"/local/#soumith-chintala","name":"Soumith Chintala","summary":"Co-founder of PyTorch, the framework behind ExecuTorch's on-device AI stack","keywords":"Thinking Machines NYU Meta"},{"type":"person","slug":"stella-biderman","href":"/local/#stella-biderman","name":"Stella Biderman","summary":"Executive Director of EleutherAI; first author of the Pythia open model suite","keywords":"EleutherAI"},{"type":"person","slug":"teknium","href":"/local/#teknium","name":"Teknium","summary":"Co-founder of Nous Research and first-listed author of the Hermes 3 and Hermes 4 technical reports","keywords":"Nous Research"},{"type":"person","slug":"thomas-wolf","href":"/local/#thomas-wolf","name":"Thomas Wolf","summary":"Co-founder and Chief Science Officer of Hugging Face; created the Transformers library","keywords":"Hugging Face"},{"type":"person","slug":"tianqi-chen","href":"/local/#tianqi-chen","name":"Tianqi Chen","summary":"Started MLC LLM, a compiler-based engine for running language models on GPUs, phones, and browsers","keywords":"Carnegie Mellon University NVIDIA"},{"type":"person","slug":"tim-baek","href":"/local/#tim-baek","name":"Tim Baek","summary":"Creator and founder of Open WebUI, a self-hosted AI interface for local and cloud models","keywords":"Open WebUI"},{"type":"person","slug":"tim-dettmers","href":"/local/#tim-dettmers","name":"Tim Dettmers","summary":"Creator and maintainer of bitsandbytes; Assistant Professor at CMU and Research Scientist at Ai2","keywords":"Carnegie Mellon University Ai2 (Allen Institute for AI)"},{"type":"person","slug":"timothy-carambat","href":"/local/#timothy-carambat","name":"Timothy Carambat","summary":"Founder of Mintplex Labs, maker of AnythingLLM, an AI application that runs locally by default","keywords":"Mintplex Labs"},{"type":"person","slug":"tri-dao","href":"/local/#tri-dao","name":"Tri Dao","summary":"First author of FlashAttention and co-author of Mamba; Princeton professor and Together AI co-founder","keywords":"Princeton University Together AI"},{"type":"person","slug":"woosuk-kwon","href":"/local/#woosuk-kwon","name":"Woosuk Kwon","summary":"Co-creator and lead maintainer of vLLM; co-founder and CTO of Inferact","keywords":"Inferact vLLM project"},{"type":"person","slug":"xuan-son-nguyen","href":"/local/#xuan-son-nguyen","name":"Xuan-Son Nguyen","summary":"Software engineer at Hugging Face and core maintainer of llama.cpp","keywords":"Hugging Face"},{"type":"person","slug":"yagil-burowski","href":"/local/#yagil-burowski","name":"Yagil Burowski","summary":"LM Studio team member; made the first commits to LM Studio's lms CLI and MLX engine repositories","keywords":"Element Labs (LM Studio)"},{"type":"person","slug":"ying-sheng","href":"/local/#ying-sheng","name":"Ying Sheng","summary":"SGLang co-author and codeowner; first author of FlexGen, for running large models on a single GPU","keywords":"LMSYS SGLang project xAI"},{"type":"place","slug":"california","href":"/places/california/","name":"California","summary":"Organizations headquartered in California","keywords":"state headquarters place"},{"type":"place","slug":"colorado","href":"/places/colorado/","name":"Colorado","summary":"Organizations headquartered in Colorado","keywords":"state headquarters place"},{"type":"place","slug":"delaware","href":"/places/delaware/","name":"Delaware","summary":"Organizations headquartered in Delaware","keywords":"state headquarters place"},{"type":"place","slug":"district-of-columbia","href":"/places/district-of-columbia/","name":"District of Columbia","summary":"Organizations headquartered in District of Columbia","keywords":"state headquarters place"},{"type":"place","slug":"florida","href":"/places/florida/","name":"Florida","summary":"Organizations headquartered in Florida","keywords":"state headquarters place"},{"type":"place","slug":"georgia","href":"/places/georgia/","name":"Georgia","summary":"Organizations headquartered in Georgia","keywords":"state headquarters place"},{"type":"place","slug":"idaho","href":"/places/idaho/","name":"Idaho","summary":"Organizations headquartered in Idaho","keywords":"state headquarters place"},{"type":"place","slug":"illinois","href":"/places/illinois/","name":"Illinois","summary":"Organizations headquartered in Illinois","keywords":"state headquarters place"},{"type":"place","slug":"maryland","href":"/places/maryland/","name":"Maryland","summary":"Organizations headquartered in Maryland","keywords":"state headquarters place"},{"type":"place","slug":"massachusetts","href":"/places/massachusetts/","name":"Massachusetts","summary":"Organizations headquartered in Massachusetts","keywords":"state headquarters place"},{"type":"place","slug":"new-jersey","href":"/places/new-jersey/","name":"New Jersey","summary":"Organizations headquartered in New Jersey","keywords":"state headquarters place"},{"type":"place","slug":"new-mexico","href":"/places/new-mexico/","name":"New Mexico","summary":"Organizations headquartered in New Mexico","keywords":"state headquarters place"},{"type":"place","slug":"new-york","href":"/places/new-york/","name":"New York","summary":"Organizations headquartered in New York","keywords":"state headquarters place"},{"type":"place","slug":"north-carolina","href":"/places/north-carolina/","name":"North Carolina","summary":"Organizations headquartered in North Carolina","keywords":"state headquarters place"},{"type":"place","slug":"oregon","href":"/places/oregon/","name":"Oregon","summary":"Organizations headquartered in Oregon","keywords":"state headquarters place"},{"type":"place","slug":"pennsylvania","href":"/places/pennsylvania/","name":"Pennsylvania","summary":"Organizations headquartered in Pennsylvania","keywords":"state headquarters place"},{"type":"place","slug":"tennessee","href":"/places/tennessee/","name":"Tennessee","summary":"Organizations headquartered in Tennessee","keywords":"state headquarters place"},{"type":"place","slug":"texas","href":"/places/texas/","name":"Texas","summary":"Organizations headquartered in Texas","keywords":"state headquarters place"},{"type":"place","slug":"washington","href":"/places/washington/","name":"Washington","summary":"Organizations headquartered in Washington","keywords":"state headquarters place"}]}