Thomson Reuters launches its own legal and tax model

Thomson Reuters has launched Thomson, its first proprietary large language model, developed in-house for legal, tax and other professional work. The company said it started from a strong open-source foundation and invested $40 million in talent and compute to train the model, rather than pursuing the much higher infrastructure spending associated with typical frontier models.

Thomson was trained using proprietary content and expertise from Westlaw, Practical Law, Checkpoint and Reuters, with subject matter experts involved from training design through final evaluations. Thomson Reuters said early testing places the model on par with recent frontier models across a range of tasks, with gains in instruction following and domain-specific reasoning. The model has been trained on less than 10% of Thomson Reuters content so far.

The company is positioning Thomson around AI sovereignty, including control over how the model is trained, where it runs and how customer information is protected. Its first deployment will be in Tabular Analysis within CoCounsel Legal for high-volume structured document review, while CoCounsel Legal will remain multi-model. A small open-weight version is also being made available on Hugging Face for academic and non-commercial validation.