Trinity-Large-Thinking

Trinity-Large-Thinking is a frontier open reasoning model designed for complex, long-horizon agents and multi-turn tool calling, released under the Apache 2.0 license. It is intended for developers, startups, and enterprises seeking reliable, cheap, and high-quality agents. The model has been trained to improve upon its predecessor, Preview, with enhanced capabilities in multi-turn tool use, context coherence, and instruction following across long-horizon agent runs.

Trinity-Large-Thinking achieves its capabilities through a combination of pretraining on 2048 NVIDIA B300s and post-training on 1152 H100s, with its inference stack powered by NVIDIA Dynamo, Blackwell Ultra GPUs, and vLLM. This optimization enables the model to score #2 on PinchBench, a benchmark measuring model capability on tasks relevant to agents like OpenClaw, while being roughly 96% cheaper than competitors at $0.90 per million output tokens on its API.

Developers and enterprises seeking to integrate high-quality, open-source reasoning models into their applications will derive the most value from Trinity-Large-Thinking. Its ability to handle complex, long-horizon tasks while maintaining coherence and following instructions makes it particularly suited for applications where reliability and efficiency are paramount. With its release under the Apache 2.0 license, Trinity-Large-Thinking offers a compelling solution for those looking to leverage open-source technology in their AI projects.