Arcee Open-Sources Trinity-Large-Thinking AI Reasoning Model

U.S. AI startup Arcee has launched Trinity-Large-Thinking, targeting agent applications that require multi-step planning and tool calls. The model uses a sparse mixture-of-experts architecture with about 398 billion parameters in total, while activating only about 13 billion parameters per token to balance reasoning performance with computational efficiency. Its open weights allow companies to deploy, modify and use the model commercially.

Data released on April 2, 2026, showed Trinity-Large-Thinking scoring 91.9 on PinchBench, close to Anthropic’s Claude Opus 4.6 at 93.3. It scored 88.0 on Tau2-Airline, outperforming the comparison models. Arcee’s API charges $0.90 per million output tokens, which the company says is about 96% cheaper than Opus 4.6.