Arcee AI Releases Trinity-Large-Thinking, a 399-Billion Parameter Open Source Reasoning Model

Arcee AI has released Trinity-Large-Thinking, a 399-billion parameter mixture-of-experts AI model published under the Apache 2.0 open source license, making it freely available for commercial use and modification, VentureBeat reported. The model is notable both for its scale and its American provenance. At a time when the most capable open-weight models have largely come from Chinese labs such as DeepSeek and Alibaba's Qwen team, Trinity-Large-Thinking represents one of the more powerful US-developed open source AI releases. VentureBeat described it as a rare example of a powerful US-made open model. Trinity-Large-Thinking is a reasoning model, meaning it is trained to work through problems step by step before arriving at a final answer -- a design pattern popularized by OpenAI's o1 and subsequently adopted across the industry. The mixture-of-experts architecture allows the model to be more computationally efficient than a dense model of equivalent parameter count by activating only a subset of parameters for any given input. Arcee AI has previously focused on model personalization and enterprise fine-tuning. The Trinity release represents a more ambitious step into frontier-scale model development. The Apache 2.0 license is among the most permissive available, allowing organizations to use, modify, and distribute the model for any purpose without royalty obligations. That stands in contrast to some other open-weight releases that include commercial use restrictions. The combination of scale, reasoning capability, and permissive licensing makes Trinity-Large-Thinking a significant release for enterprises evaluating open source AI deployments.