Model details
Trinity Large Thinking is a 398-billion parameter sparse Mixture-of-Experts model built on the AfmoeForCausalLM architecture. Designed specifically for agentic tasks, it excels at multi-step planning, complex tool calling, and long-horizon workflows. The model distinguishes itself by generating explicit reasoning traces within dedicated tags, allowing users to observe its internal logic before it produces a final response. This architecture is optimized to maintain coherence across multi-turn conversations, making it a robust choice for developers building sophisticated autonomous agents.
The model is the result of an intensive two-month scaling effort focused on refining supervised fine-tuning and reinforcement learning pipelines to align with the base model's capacity. By prioritizing sovereignty and auditability, it offers a high-performance, downloadable alternative for enterprises seeking to avoid vendor lock-in. Its design emphasizes practical utility in demanding environments, delivering strong results on benchmarks like LiveCodeBench and τ²-Bench. As an open-weight solution, it provides the flexibility for organizations to host and customize their own reasoning infrastructure for specialized, high-stakes applications.
Transparent token rates
Rates are shown per one million tokens. Combined means one million input plus one million output tokens.
OpenRouter
Arcee, a 26-person U.S. startup, released Trinity Large Thinking, a 400B-parameter open-weight reasoning model. CEO Mark McQuade positions the model as a Western alternative to Chinese open models, enabling on-premises use and API access. While Arcee acknowledges it does not yet outperform closed-source systems from ma
OpenRouter
Compare Qwen3.5 Plus 2026-04-20 from Qwen and Trinity Large Thinking from arcee-ai on key metrics including benchmarks, price, context length, and other model features.
This exact model name is also listed by 3 other providers.
