Thomson Reuters builds AI model using Chinese tech to reduce reliance on Claude

Key facts

• Thomson Reuters has launched its own AI model named Thomson-1.
• The model is built upon Alibaba's open-source Qwen model, adapted by Thomson Reuters and Imperial College London.
• This initiative aims to decrease the company's dependence on expensive third-party AI services, such as Anthropic's Claude.
• Thomson-1 will be utilized for tasks like document review, leveraging the company's specific expertise.
• The development signifies a trend of Western companies utilizing Chinese open-source AI technology.

Thomson Reuters has introduced its proprietary AI model, Thomson-1, which is built upon Alibaba's open-source Qwen model. This strategic move aims to reduce the company's reliance on costly AI solutions from providers like Anthropic, while simultaneously leveraging its own expertise, particularly in document review.

Chief Technology Officer Joel Hron stated that Thomson-1 was developed by adapting the Qwen model over several months by a joint team from Thomson Reuters and Imperial College London. The company views this in-house development as a way to build long-term value, akin to buying a house rather than renting.

The decision comes amid rising costs associated with large AI models and chatbots. The use of Chinese open-source AI by Western companies is becoming more common, though it has drawn scrutiny. Anthropic has accused Chinese labs of illicitly using its model outputs, and U.S. Senator Tom Cotton has voiced concerns about potential security risks associated with Western companies using Chinese open-source models.

Hron clarified that Thomson-1 is not intended to completely replace Thomson Reuters' work with Anthropic, especially for its AI legal assistant, CoCounsel, which still relies heavily on Claude. However, the company's objective is for Thomson-1 to power an increasing number of CoCounsel's capabilities over time.