Trinity Mini
Release in the Trinity family · version Mini
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.1
- Model hub: Model card (external site: huggingface.co)
- License: License (OpenMDW-1.1) (external site: huggingface.co)
- Release notes: Release announcement (external site: arcee.ai)
- Paper: Technical report (external site: arxiv.org)
Availability and license
Overall availability
Weights download from Hugging Face without an access gate. Also served through Arcee's API and OpenRouter.16
Availability is separate from permission: read the license before using or redistributing.
Arcee released Trinity Mini on 2025-12-01 under Apache 2.0. On 2026-05-28 the Hugging Face repository changed its license to OpenMDW-1.1 as part of Arcee's move of all Trinity releases to that license. OpenMDW-1.1 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
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.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Weights are in the ungated Hugging Face repository; a GGUF conversion is published as a separate repository.1 |
| Inference codeIs code for running the model published? | Public | The repository includes modeling code (modeling_afmoe.py), and the model card documents Transformers, vLLM, llama.cpp, and LM Studio.14 |
| Training codeIs the code used to train the model published? | Unknown | The technical report says training used a modified version of TorchTitan but does not link a release.8 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The technical report describes the 10-trillion-token, three-phase pretraining mix for Trinity Nano and Mini. No public release of the data itself was found.8 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The technical report documents architecture and optimizer choices for the family; post-training details for Mini are limited.8 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The model card shows a benchmark chart and the technical report includes results; evaluation code is not linked.18 |
What it is useful for
Arcee presents Trinity Mini as the mid-sized model in the family, for cloud or on-premises deployment.5
Run and use notes
- The model card documents support in vLLM 0.11.1, llama.cpp release b7061, LM Studio, and Transformers (main branch, or a released version with trust_remote_code enabled).1
Organization context
Provenance and derivatives
Post-trained from Arcee's Trinity-Mini-Base checkpoint. The model card says the training data was gathered and curated with Datology and training ran on a cluster provided by Prime Intellect.1
- Derived from: Trinity-Mini-Base (external site: huggingface.co) — Same organization; pretrained base checkpoint.
Other releases in the Trinity family
- Trinity-Large-ThinkingModel-disclosure tier (USASI rubric v0.1): Open-weight
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.