Trinity Large Thinking

Reasoning-optimized variant of Trinity Large (398B / 13B active MoE) with extended chain-of-thought thinking before responding. Post-trained for long-horizon agentic tasks: multi-turn tool calling, sustained context coherence, clean instruction following, and stable behavior across long-running agent loops. Relicensed to OpenMDW-1.1 (Linux Foundation) on 2026-05-29, from Apache 2.0.

AA Intelligence Index v4.3: 11 (above the open-weight median for models of similar size). 132 tok/s output speed, 0.91s time-to-first-token. tau2-Airline: 88.0%, AIME25: 96.3%, SWE-Bench: 63.2%, PinchBench: 91.9% (#2 globally). #1 most-used open model in the US on OpenRouter. Available from Arcee AI and Parasail (FP8).