U.S. startup Arcee, a company with 26 employees, has introduced its new reasoning model, Trinity Large Thinking. The model follows the company’s previous work in developing a significant 400-billion-parameter open-source large language model, reportedly on a budget of $20 million.
Arcee CEO Mark McQuade stated that Trinity Large Thinking represents the most capable open-weight model ever released by a non-Chinese company. This release aligns with Arcee’s objective to provide U.S. and Western companies with robust alternatives to models originating from China.
The company aims to address perceived risks associated with using Chinese-developed models, which some fear could lead to data or power being held by a government with differing ideals from the Western world. Arcee’s models allow companies to download, customize, and deploy them on their own premises, or access a cloud-hosted version via an API.
While Arcee’s models do not currently surpass the performance of proprietary, closed-source models from major labs like Anthropic or OpenAI, they offer an independent solution. The company highlights that its models are not subject to the evolving policies of larger providers, citing instances where access or pricing for other models has changed for users of tools like OpenClaw.
Data from OpenRouter indicates that Arcee’s model has become a leading choice among models utilized with OpenClaw. Benchmarks shared by Arcee suggest Trinity Large Thinking is comparable in capability to other prominent open-source models.
Furthermore, Arcee’s Trinity models are released under the Apache 2.0 license, which is considered a highly open standard for open-source licenses. This distinguishes them from some other U.S.-built open models, such as Meta’s Llama 4, which has faced questions regarding its licensing terms.
Source: Original
