Startup Arcee launches new language model Trinity Large Thinking

Arcee, a small 26-person U.S. startup, has developed a new language model called Trinity Large Thinking with 400 billion parameters on a budget of $20 million. CEO Mark McQuade claims it is the most powerful open model ever released by a non-Chinese company. Arcee's main goal is to provide U.S. and Western companies with an alternative to Chinese models, which are considered risky due to potential data sharing with a government that does not share Western ideals.

Companies can download the Trinity model, customize it to their needs, and use it on their premises. A cloud-hosted version is also available via API. While Arcee's models do not outperform the closed-source models from major labs like Anthropic or OpenAI, they are not at the mercy of those companies' decisions.

For instance, Anthropic recently announced that subscriptions would no longer cover OpenClaw usage, forcing users to pay extra. In contrast, Arcee reports that their models are gaining popularity among OpenClaw users, as evidenced by data from OpenRouter.

As for the Trinity Large Thinking model itself, benchmark results indicate it is comparable to other leading open-source models. While it is not a direct competitor to Meta's Llama 4, it avoids the licensing issues associated with Meta's model. All of Arcee's models are released under the Apache 2.0 license, making them accessible to a wider audience.

Among the many startups offering open-source models, Arcee stands out with its innovative approach and commitment to excellence, garnering interest and support from the community.