Thomson Reuters Builds Its Own AI Model To Own The Legal Stack

Thomson Reuters has launched its own proprietary AI model, and the move says something important about where enterprise AI is going. Companies with valuable data do not want to rent intelligence forever.

In an official announcement, Thomson Reuters said its proprietary LLM, called Thomson, was developed in-house, remains fully owned and controlled by the company, and is trained around its professional data assets across legal, tax, compliance, risk and news.

The company says Thomson was trained and can run at a fraction of the cost of comparable frontier models. That is a significant claim because many enterprise AI deployments are now running into the same problem: general frontier models are powerful, but they can be expensive to use at scale and may not know enough about a specialist domain without extra grounding.

Thomson Reuters has a different advantage. It owns legal and professional content, expert taxonomies, editorial processes and products such as Westlaw, Practical Law and CoCounsel. That gives it the kind of proprietary context a general AI lab cannot easily copy.

This is not about replacing every outside model overnight. Thomson Reuters already works with partners such as Anthropic, and we recently covered Thomson Reuters bringing Claude agents into legal drafting. The new model gives the company more control over the parts of the stack where its own content and workflows matter most.

That control has strategic value. If a company owns the model, the training data, the evaluation process and the product surface, it can optimize for its own customers rather than wait for a frontier lab to prioritize legal, tax or compliance use cases.

It also reduces cost risk. As AI usage grows inside legal and professional workflows, inference costs can become material. A smaller specialist model that performs very well on targeted tasks may be more useful than a larger general model that is expensive and less grounded in professional content.

There is a wider lesson here for enterprise software companies. The future may not be one giant model serving every industry. It may be a mix of frontier models, open-weight foundations and proprietary specialist models trained on trusted company data.

We also wrote recently about Wolters Kluwer turning legal content into AI-ready intelligence. Thomson Reuters is moving in the same direction from another angle: not only structuring professional data for AI, but owning a model that can use it.

The result is a clearer picture of enterprise AI’s next phase. The companies with deep domain content are no longer waiting to be disrupted by AI labs. They are building their own AI infrastructure around the knowledge they already control.