Thomson Reuters Spends $40 Million to Build Its Own Professional AI Model

Thomson Reuters has launched its first proprietary large language model as the professional-information company moves from integrating third-party artificial intelligence to owning more of the intelligence layer itself.\n\nThe model, called Thomson, was developed from an open-source foundation and specialized using Thomson Reuters proprietary legal, tax and professional content. The company says it invested approximately $40 million in talent and computing to develop the system.\n\nThomson Reuters says it fully owns and controls the resulting model, giving it greater control over inference costs, privacy, model behavior and future deployment.\n\nThe first production use will be inside the Tabular Analysis capability of CoCounsel Legal, where the model will handle high-volume structured document review. CoCounsel will remain a multi-model product and can continue using other leading models when they are better suited to a task.\n\nThe company says less than 10% of its proprietary content has been incorporated into Thomson so far.\n\nThomson Reuters is also releasing a smaller open-weight version of the model on Hugging Face for academic and non-commercial evaluation.\n\nThe launch points toward a potentially important shift in enterprise AI: companies with valuable proprietary datasets may increasingly specialize open models rather than relying exclusively on general-purpose frontier-model providers.