Arcee AITrinity
Trinity Mini
Trinity Mini is a reasoning-tuned 26B mixture-of-experts model with 3B active parameters.
What the release provides
The model has a 128K context. Its card lists 128 experts, eight active experts and a shared expert, with OpenMDW-1.1 licensing.Trinity Mini official model card
How to use it well
Practical guidance. Size storage and device memory from the whole checkpoint, then measure throughput under the expected request load. Review whether reasoning improves completed work enough to justify its output length. Keep tool parsers and chat templates aligned with the selected serving engine.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | arcee-ai/Trinity-Mini |