I can’t help rooting for tiny open source AI model maker Arcee

The U.S. startup Arcee has released Trinity Large Thinking, a new 400B-parameter open-weight model positioned as a direct response to the market's reliance on Chinese-based alternatives. CEO Mark McQuade asserts it is the “most capable open-weight model ever released by a non-Chinese company,” aiming to give Western companies an option that mitigates perceived risks associated with data and governance tied to non-Western governments.

Developed by a 26-person team on a modest $20 million budget, Arcee’s models stand out for their accessibility and licensing. Unlike competitors such as Meta’s Llama 4, which has a more restrictive license, Trinity is released under the permissive Apache 2.0 license. This allows companies to download, customize, and run the model on their own infrastructure for maximum control, though a cloud-hosted API version is also available for ease of use.

Arcee's market entry highlights a growing demand for independence from the unpredictable policies of large, closed-source AI labs. While not outperforming models from Anthropic or OpenAI on raw benchmarks, Arcee offers stability. This was underscored when Anthropic recently altered its subscription terms, forcing users of the open-source agent tool OpenClaw to pay extra. Subsequently, Arcee's model has reportedly become a popular choice within the OpenClaw community, illustrating the appeal of a model that isn't subject to the strategic whims of a major provider.

Arcee's strategic value is not in topping leaderboards, but in offering enterprise-grade 'model sovereignty.' By providing a capable, permissively licensed, and self-hostable foundation model, it directly addresses corporate demand for stability and insulation from the sudden platform shifts and geopolitical uncertainties that characterize the current AI market.