Trinity-Large-Thinking: Agentic Reasoning MoE

Trinity-Large-Thinking is a 398B-parameter sparse Mixture-of-Experts (MoE) model from Arcee AI, featuring approximately 13B active parameters per token. It is specifically optimized for reasoning and agentic workflows, built upon the Trinity-Large-Base and further post-trained with extended chain-of-thought reasoning and agentic Reinforcement Learning (RL).

Key Capabilities & Features

• Agentic-first design: Purpose-built for tool calling, multi-step planning, and complex agent workflows.
• Native Reasoning Traces: Generates explicit internal reasoning within <think>...</think> blocks, which are crucial for its performance and must be maintained in context for multi-turn conversations and agentic loops.
• State-of-the-art Agentic Performance: Achieves 94.7% on τ²-Bench, 91.9% on PinchBench, and 98.2% on LiveCodeBench, demonstrating strong capabilities in agentic tasks.
• Broad Compatibility: Works out-of-the-box with major agent frameworks like OpenClaw and Hermes Agent.
• Sparse MoE Architecture: Utilizes 256 experts with 4 active experts per token, and a 512k context length after extension.

Ideal Use Cases

Trinity-Large-Thinking is best suited as the reasoning backbone for AI agent systems, including applications requiring:

• Tool-calling and function execution.
• Multi-step planning and task completion.
• Complex agentic loops where explicit reasoning traces enhance reliability and performance.