Thomson Reuters launches Thomson-1, its first in-house AI model built on Alibaba's Qwen

Thomson Reuters launched Thomson-1, its first in-house AI model, on Monday, built on a realigned version of Alibaba's open-source Qwen as the Canadian data and legal giant moves to reduce its reliance on costly AI from Anthropic.

The model is based on Snowdon, an intermediate model built over several months by a joint team from Thomson Reuters and Imperial College London that "realigned" Qwen, chief technology officer Joel Hron told Business Insider. The team made sure it was "ethically and politically de-biased and safe to use," he said.

Thomson-1 will take over some tasks previously handled by Claude, although Hron said it is not intended to replace the company's work with Anthropic and other labs. It will start with document review, an area where Thomson Reuters has deep expertise.

The shift comes as large companies feel the pinch from high AI costs driven by chatbots like Claude and OpenAI's Codex. Hron cited cost as one of the main reasons for building the model in-house. Owning the model lets the company build on its own intellectual property instead of continually paying outside AI firms; he compared the difference to renting versus buying a house.

Thomson Reuters expanded its partnership with Anthropic in May for CoCounsel, its AI legal assistant, which Hron said still relies mostly on Claude. "Our main objective is to make Thomson the model that powers more and more of CoCounsel's capabilities over time," he told Business Insider.

The bet on Chinese open source is not without controversy. Anthropic has accused Chinese labs of illicitly "distilling" its model outputs to train their latest chatbots, and Senator Tom Cotton has raised security concerns, including backdoors, about US companies such as Airbnb and Cursor using Chinese open-source models. Hron, however, said "there's nothing that necessarily ties us to Qwen."

The launch signals that Western enterprises are increasingly treating Chinese open-source models as a practical route to cheaper, self-owned AI. The key thing to watch is how quickly Thomson-1 takes over more of CoCounsel's workload, and whether other professional-services firms adopt a similar own-model-on-open-source-base playbook.