USASI
Model release

Trinity-Large-Thinking

Release in the Trinity family · version Large-Thinking

Maintained by Arcee AI1

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.1

Last reviewedEntry updated Documented release Apr 1, 2026

Availability and license

Overall availability

Public

Weights download from Hugging Face without an access gate. Also served through Arcee's paid API and OpenRouter.16

Availability is separate from permission: read the license before using or redistributing.

Arcee announced this release on 2026-04-01 under Apache 2.0. On 2026-05-28 the Hugging Face repository changed its license to OpenMDW-1.1, and Arcee states it moved earlier Trinity releases to that license. OpenMDW-1.1 grants broad use rights and places no restrictions on outputs. It requires keeping the license and notices when redistributing, and ends the rights of anyone who sues claiming the materials infringe a patent or copyright.6372

Model-disclosure tier

Computed from the checklist below using USASI rubric v0.1. An editorial category, not a certification.
Model-disclosure tier (USASI rubric v0.1): Open-weight

The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.

The weights are under a license that is not on the rubric's OSI-approved list. Read its terms before use.

How tiers are computed

Public materials checklist

Items for a model under USASI rubric v0.1. Unknown means unassessed or insufficient evidence.
Public materials checklist for Trinity-Large-Thinking
ItemStatusNotes and evidence
WeightsCan the general public download the model parameters for this release?PublicWeights are in the ungated Hugging Face repository; Arcee also publishes GGUF, FP8, NVFP4, and W4A16 variants as separate repositories.15
Inference codeIs code for running the model published?PublicThe repository includes modeling code (modeling_afmoe.py), and the model card documents running it with vLLM and Transformers.14
Training codeIs the code used to train the model published?UnknownThe technical report says training used a modified version of TorchTitan but does not link a release.9
Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access.PartialThe technical report describes the pretraining mix (curated web, synthetic, multilingual, and code data). No public release of the data itself was found.9
Training recipeAre the training configuration and procedure documented in enough detail to follow?PartialPretraining settings are documented in the technical report. The post-training for this variant is described only briefly on the model card.91
Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only.PartialBenchmark results are published on the model card; evaluation code is not linked.1

What it is useful for

The model card documents it for tool calling, multi-step planning, and use as the model behind agent frameworks, and notes that reasoning content must be kept in the conversation history across turns.1

Run and use notes

Documented facts only. No hardware or performance claims are made without a cited source and stated assumptions.
  • The model card documents serving with vLLM 0.11.1 or later, using the deepseek_r1 reasoning parser and qwen3_coder tool-call parser, and running with Transformers from the main branch or with trust_remote_code enabled.1

Organization context

Provenance and derivatives

Post-trained from Trinity-Large-Base, Arcee's own 17-trillion-token pretrained checkpoint. The model card names Datology as data partner and Prime Intellect as compute partner.1

Other releases in the Trinity family

  • Trinity MiniModel-disclosure tier (USASI rubric v0.1): Open-weight

Trinity family overview

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S. headquarters

Developed and published by Arcee AI, Inc., a U.S. company with a documented San Francisco base (see the arcee-ai record). It is post-trained from Arcee's own Trinity-Large-Base checkpoint.18

Assessed Sep 29, 2026

Sources

  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
    Trinity-Large-Thinking: Scaling an Open Source Frontier Agent (external site: arcee.ai)

    Arcee AI · Announcement · published Apr 1, 2026 · accessed Sep 29, 2026

  7. 7.
    Trinity is moving to OpenMDW-1.1 (external site: arcee.ai)

    Arcee AI · Announcement · published May 29, 2026 · accessed Sep 29, 2026

  8. 8.
    The Trinity Manifesto (external site: arcee.ai)

    Arcee AI · Announcement · published Dec 1, 2025 · accessed Sep 29, 2026

  9. 9.
    Arcee Trinity Large Technical Report (external site: arxiv.org)

    arXiv · Paper · published Feb 19, 2026 · accessed Sep 29, 2026

This listing is not an endorsement, a safety assessment, or a federal approval.

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