Called "Thomson" and built on an open basis for $40 million, the model is fed by Westlaw, Practical Law, Checkpoint and Reuters, and deployed in CoCounsel Legal.
On August 24, 2026, Thomson Reuters announced the launch of Thomson, its first major proprietary language model. The group did not start from scratch: it built on an open source foundation, then led the model over decades of home content — Westlaw, Practical Law, Checkpoint and Reuters dispatches — with the input of hundreds of specialists in the field to guide the result.
The announced investment is $40 million, including calculation and recruitment. This is modest in terms of the budgets of the border laboratories, and the company makes one argument: the model trains and runs for a fraction of the cost of its generalist equivalents, while remaining entirely owned and controlled by the group. Specialisation is the bet: law, taxation, accounting, compliance.
Thomson was first put to work in CoCounsel Legal, the group's legal assistant, with the promise of additional capabilities and sovereign AI options later. Logic — a publisher sitting on proprietary archives who decides to train his model rather than hire someone else's — is becoming an increasingly common strategy among holders of specialized documentary holdings.
Read the full article on the original source
Go to the source →