Thomson Reuters has launched Thomson, the company's first proprietary large language model — developed in-house for $40 million, a fraction of what frontier AI labs typically spend. The company says the model performs on par with leading frontier models across professional legal and tax tasks.
Rather than starting from scratch, Thomson Reuters built Thomson on a strong open-source foundation and then applied state-of-the-art mid-training and post-training techniques using decades of proprietary content from Westlaw, Practical Law, Checkpoint, and Reuters. Hundreds of subject matter experts were integrated from the design of training objectives through final evaluations.
"For years, the AI industry has treated scale as the answer: bigger models, more compute, more money. Thomson shows there is another path," said Joel Hron, Chief Technology Officer.
Early evaluations put Thomson on par with the latest frontier models across a range of tasks. The model has been trained on less than 10% of Thomson Reuters' total content, suggesting significant room for further improvement. The company describes Thomson as "Fiduciary-Grade" AI — designed for professionals with duties of care and accountability where accuracy is paramount.
The model's first deployment is inside Tabular Analysis in CoCounsel Legal, where it processes structured document reviews. Thomson Reuters plans to extend it across its legal and tax portfolio with additional sovereign AI options. The company is also releasing a "small" version as an open-weight model on Hugging Face for academic use.
Independent evaluations from law professors at Washington University and Cornell's Legal AI Lab found Thomson's citation quality competitive with ChatGPT and Claude, with Thomson offering advantages in domain-specific reasoning and transparency.
