Arcee AI Open-Sources Trinity-Large-Thinking, a 400B Parameter Reasoning Model
Arcee AI has released Trinity-Large-Thinking, a 400-billion-parameter open-source reasoning model licensed under Apache 2.0, filling a gap in the U.S.-made open-source AI landscape that has widened as Chinese labs have retreated and Meta’s Llama releases have slowed.
The model is available for download and commercial use, making it one of the most powerful openly licensed reasoning models released by a U.S.-based company to date.
Contents
show
Architecture: Massive Scale, Efficient Inference
Trinity-Large-Thinking uses a sparse Mixture-of-Experts (MoE) architecture. While the model has 400 billion total parameters, only approximately 13 billion are active during any given inference pass. This design delivers high reasoning capability without the prohibitive compute costs associated with dense models of comparable size.
The result is a model that can compete with proprietary alternatives on reasoning benchmarks while remaining economically viable to run on enterprise-grade hardware. Organizations with access to multi-GPU inference infrastructure can deploy Trinity at costs significantly below what frontier proprietary models typically require.
Filling a Strategic Gap
Arcee positioned the release explicitly as a response to the thinning supply of powerful, openly licensed U.S.-made models. Several Chinese AI labs that previously contributed heavily to the open-source ecosystem have reduced their public model releases amid geopolitical pressures. Meta, long a major contributor through its Llama series, has also moderated its release cadence as it evaluates commercial and safety considerations.
Trinity-Large-Thinking arrives at a moment when enterprises increasingly want open-weight models they can fine-tune, audit, and deploy without vendor lock-in. The Apache 2.0 license, which permits commercial modification and redistribution, gives Trinity broad applicability across industries with strict data governance requirements.
Enterprise and Research Appeal
The model’s combination of scale, efficiency, and permissive licensing makes it attractive to both enterprise and research users. Teams that previously relied on API access to proprietary reasoning models for tasks like complex document analysis, code generation, or multi-step planning can now explore self-hosted alternatives with greater control over data and cost.
Arcee says the model has been evaluated across standard reasoning benchmarks and achieves competitive scores against proprietary models from major labs, though independent evaluations are still emerging from the research community.
The release adds meaningful weight to the argument that the open-source AI ecosystem can produce frontier-class models without depending on a small number of large foundation labs.