Arcee Trinity-Large-Thinking: 400B Open-Source MoE Hits #2 on PinchBench at 96% Less Than Opus 4.6

Arcee AI·high signal

Arcee AI released Trinity-Large-Thinking under Apache 2.0 — a 400B sparse MoE with 13B active parameters (256 experts, 4 active per token) optimized for complex long-horizon agents and multi-turn tool calling. It ranks #2 on PinchBench behind only Opus 4.6 while costing ~$0.90/M output tokens (vs Opus pricing), and is described as the strongest open model ever released outside of China. Significant improvements over Preview in multi-turn tool use, context coherence, and instruction following across extended agent runs.