Thomson Reuters built its own AI model off Chinese tech to rely less on Claude

Thomson Reuters has unveiled its first AI model, Thomson-1, constructed using a realigned version of Alibaba's Qwen3.5 model, with chief technology officer Joel Hron revealing the news on Monday. The Canadian data and legal giant is aiming to minimize reliance on costly AI from firms like Anthropic, and opted for the Chinese open-source model to achieve this goal.

Thomson-1 will handle some tasks currently managed by Claude, but Hron emphasized that it is not meant to replace the company's collaboration with Anthropic and other labs, but rather focus on areas where Thomson Reuters has expertise, beginning with document review. The shift towards building an in-house model comes as large companies face the strain of high AI costs, driven by chatbots such as Claude and OpenAI's Codex.

Hron explained that building its own model allows Thomson Reuters to leverage the company's expertise in specific areas and save costs in the process. Despite the rise of Chinese AI models, the company has continued to partner with Anthropic for its AI legal assistant, CoCounsel, which still relies predominantly on Claude. Hron highlighted that their primary goal is to enhance Thomson Reuters' capabilities through its own model, rather than solely relying on external AI providers.

The development of Chinese AI models has not been universally welcomed, with Anthropic alleging that Chinese labs illicitly distill outputs from their AI models to train their chatbots, prompting calls for US restrictions on such practices. Other US leaders have also expressed concerns about security threats and potential backdoors associated with the use of Chinese open-source models.