Thomson Reuters announced the launch of Thomson, its first proprietary large language model developed in-house, on August 24, 2026. According to the company, Thomson was trained and operates at a fraction of the cost of comparable frontier models and remains fully owned and controlled by Thomson Reuters.
The company invested $40 million to train Thomson into what it describes as specialized intelligence for professional work, focusing on talent and compute. Rather than following the typical path of frontier AI labs, which spend billions on compute and years on infrastructure, Thomson Reuters started from a strong open-source foundation and specialized it using proprietary content and domain expertise.
Thomson was trained on proprietary content from Westlaw, Practical Law, Checkpoint, and Reuters, with hundreds of subject matter experts involved from training design through final evaluations. According to the company, the model has been trained on less than 10% of Thomson Reuters’ content so far. The company emphasizes that Thomson is built to what it calls “Fiduciary-Grade™ standards,” meeting requirements for professionals with duties of care and accountability.
According to early evaluations cited by the company, Thomson performs on par with the latest frontier models across a range of tasks. External academics testing the model reported competitive performance. Jonathan H. Choi of Washington University School of Law noted that Thomson’s responses compared favorably to ChatGPT and Claude in corporate tax questions, particularly for its inclusion of links to treatises. Professor Samuel Dahan of the Queen’s Conflict Analytics Lab and Cornell Legal AI Lab found Thomson’s citation quality “generally competitive with leading frontier models.”
Thomson demonstrates particular strength in domain-specific tasks. According to Thomson Reuters, the model shows “meaningful uplift” in instruction following and in navigating dense, domain-specific content. The company notes that proprietary training and human subject matter expertise applied to a strong foundation produced gains that content access alone does not achieve.
The company is emphasizing AI sovereignty as a key differentiator, addressing questions about model training, embedded behaviors and biases, where it runs, and how user information privacy is protected. Thomson Reuters is positioning sovereignty and verification as competitive advantages in professional AI.
Thomson’s first deployment is in Tabular Analysis within CoCounsel Legal for structured document review. The company plans to extend Thomson across its legal and tax portfolio with additional sovereign AI options to follow. Thomson Reuters is making a “small” version of Thomson available as an open-weight model on Hugging Face for academic and non-commercial use.
Key facts
- Thomson Reuters invested $40 million to train Thomson, its proprietary large language model
- Thomson was trained on less than 10% of the company’s proprietary content so far
- The model performs on par with leading frontier models according to early evaluations by external academics
- Thomson demonstrates particular strength in domain-specific tasks and instruction following
- The model operates at a fraction of the cost of comparable frontier models
- Thomson’s first deployment is in CoCounsel Legal’s Tabular Analysis for document review
- A small version of Thomson is available as open-weight on Hugging Face for academic use