Independent project. Not a U.S. government website.

USASI

Hub

Foundation models and the labs that build them

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.

Model developers in this catalog make their models available in different ways. Some offer them as hosted services: Anthropic's documentation lists each Claude model by its ID on the Claude API and on cloud platforms run by Amazon, Google, and Microsoft. Some publish the weights: Ai2 says its Olmo 3 models were pre-trained on the Dolma 3 dataset and post-trained on the Dolci datasets, and that it is releasing the code, checkpoints, and training details. Others do both. OpenAI serves models through its API and also publishes gpt-oss, which its model card calls OpenAI's open-weight models, under the Apache 2.0 license.1534

A model name often covers many releases. The Olmo 3 model card lists 7B and 32B models in Instruct and Think variants, and publishes a separate checkpoint for each training stage (supervised fine-tuning, preference tuning, and a final reinforcement-learning stage). In this catalog, a family record summarizes a named line of models, while licenses, availability, and openness tiers attach only to individual releases, because they can differ between releases in the same family.510

Downloadable weights come with different terms. Gemma 4 and gpt-oss use Apache 2.0. The Llama 4 Community License incorporates an acceptable use policy and requires anyone who distributes the models, or a product that contains them, to display "Built with Llama"; a model trained or improved with them and then distributed must have a name beginning with "Llama". The Llama 4 Scout repository on Hugging Face asks you to sign in and review these conditions before access. The xAI Community License for Grok 2 allows commercial use only within xAI's acceptable use policy and bars using the materials or their outputs to train other general-purpose models.64789

A permissive license does not settle every question. Ai2's Olmo 3.1 32B Think card gives an Apache 2.0 license and also says the model is intended for research and educational use in line with Ai2's Responsible Use Guidelines. The catalog keeps availability separate from permission: being publicly downloadable does not by itself make a release unrestricted, open source, commercially usable, or redistributable.510

Hosted models change on the provider's schedule. Anthropic says every Claude model ID is a pinned snapshot, that a deprecated model is given a retirement date, and that requests to a retired model fail. It gives at least 60 days' notice before retiring publicly released models on the platforms it operates, while Amazon Bedrock and Google Cloud set their own schedules. Check the lifecycle page for any hosted model you build on.12

What this hub covers

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.

Reading path

  1. 1.What open weight and open source meanStart here. Downloadable weights and permission to use them are separate questions.
  2. 2.Model familyA named line of related models released over time.
  3. 3.Model releaseOne specific published model, with its own files and its own terms.
  4. 4.Families and releasesWhy licenses, availability, and tiers attach to releases and never to a family.
  5. 5.How to read a model cardFind the license, intended uses, base model, and limitations for one release.
  6. 6.Acceptable use policyUse policies can limit applications even when the weights download freely.
  7. 7.Hosted or local?What changes when you call a hosted model instead of running downloaded weights.

In the catalog

Examples chosen to cover the subject; not a ranking or a complete list.

Primary documents

Sources · reviewed Oct 2, 2026

  1. 1.
    Models overview (external site: platform.claude.com)

    Anthropic (Claude Platform documentation) · Documentation · accessed Oct 2, 2026

  2. 2.
    Model deprecations (external site: platform.claude.com)

    Anthropic (Claude Platform documentation) · Documentation · accessed Oct 2, 2026

  3. 3.
  4. 4.
    gpt-oss-120b model card (external site: huggingface.co)

    OpenAI (Hugging Face) · Model card · accessed Oct 2, 2026

  5. 5.
    Model card for Olmo 3.1 32B Think (external site: huggingface.co)

    Ai2 (Hugging Face) · Model card · accessed Oct 2, 2026

  6. 6.
    Gemma 4 license (external site: ai.google.dev)

    Google AI for Developers · License · accessed Oct 2, 2026

  7. 7.
  8. 8.
    meta-llama/Llama-4-Scout-17B-16E (external site: huggingface.co)

    Meta (Hugging Face) · Model card · accessed Oct 2, 2026

  9. 9.
    xAI Community License Agreement (Grok 2) (external site: huggingface.co)

    xAI (Hugging Face) · License · accessed Oct 2, 2026

  10. 10.

Support Us

Help keep USASI useful.

Optional. No USASI account required. Payment takes place on the linked provider’s website (Buy Me a Coffee).

About supporting this project