Thomson Reuters launches its first in-house AI model

Thomson Reuters has launched Thomson, its first proprietary large language model. Developed in-house, the model remains under the company's control.

The launch gives Thomson Reuters its own large language model as professional services groups weigh how much to rely on third-party AI systems for legal, tax and compliance work. It also marks a shift for a company that has largely integrated outside models into products such as CoCounsel while building on its own data and editorial assets.

Executives said Thomson was trained from an open-source base and refined using the company's legal, tax and news content, including material from Westlaw, Practical Law, Checkpoint and Reuters. Thomson Reuters said it spent USD $40 million on the effort, covering talent and computing resources, rather than the much larger sums often associated with frontier model development.

So far, the model has been trained on less than 10% of the company's content base. The focus now is on further specialisation rather than simply adding more data.

Control and cost

Thomson Reuters presented the launch as an effort to build a model tailored to professional work while retaining ownership of the underlying system and avoiding the inference costs associated with many large third-party models. It said concerns about AI sovereignty, including how systems are trained, where they run, what biases they carry and how customer information is handled, are becoming more important for professional users.

It also said customer data is not used to train the model without explicit consent. That is likely to matter for law firms, corporate legal teams and tax professionals handling sensitive documents under regulatory and fiduciary obligations.

Joel Hron, Chief Technology Officer at Thomson Reuters, said the company wanted to challenge the idea that ever-larger models and spending are the only route to competitive AI systems.

"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, Thomson Reuters.

"Start with a strong foundation, specialize it deeply for the work that matters, and you can build intelligence that is highly capable, far more efficient and entirely under your control. We think that changes the economics of professional AI," Hron said.

Product rollout

The first use of Thomson will be in Tabular Analysis within CoCounsel Legal, Thomson Reuters' legal AI assistant. The feature is aimed at high-volume, structured document review for law firms and in-house legal departments.

CoCounsel will remain a multi-model product, the company said. Thomson Reuters plans to use its own model where it sees a clear advantage and continue using other leading models in other parts of the service.

It also intends to extend Thomson models across its legal and tax portfolio and add more sovereign AI options. No timetable was given.

Early testing

Before the launch, Thomson Reuters opened the model to a group of legal and AI academics for direct evaluation. It said it will continue making the model available to external parties for validation and development, and will release a small open-weight version for academic and non-commercial use.

Two outside academics cited by the company gave early views on the model's performance. One compared Thomson with ChatGPT and Claude on corporate tax questions, while another assessed citation quality on Canadian employment law questions.

"I tested Thomson against ChatGPT and Claude using some of the more challenging questions students have asked in my Corporate Tax class. All three models answered the questions correctly, but I preferred Thomson's responses overall. I especially appreciated the links to treatises, which made its responses more transparent and useful for legal work," said Jonathan H. Choi, Washington University School of Law.

"Our evaluation found Thomson's citation quality generally competitive with leading frontier models, even when tested on Canadian employment-law questions without a Canada-specific setting," said Professor Samuel Dahan, Director, Queen's Conflict Analytics Lab and Cornell Legal AI Lab.

Competitive position

Thomson Reuters said its internal evaluations place Thomson on a level with leading frontier models across a range of tasks, while showing stronger gains in instruction following and work involving dense domain-specific material. It argued that access to content alone is not enough to produce expert-level results, and that proprietary training combined with human subject-matter expertise can deliver better outcomes.

The claim speaks to a broader AI debate over whether industry-specific systems can outperform general-purpose models on specialised tasks. For publishers and information companies with large proprietary archives, Thomson Reuters' approach may test whether deep vertical data can be turned into defensible AI products rather than simply licensed to external model developers.

Steve Hasker, Chief Executive Officer of Thomson Reuters, said the company sees its long-held content base and editorial expertise as central to that effort.

"Thomson proves what's possible when you build AI on decades of proprietary content and editorial expertise," said Steve Hasker, Chief Executive Officer, Thomson Reuters.

"That's an advantage only Thomson Reuters has, and it shows in the results: our early evaluations put Thomson on par with the latest frontier models across a range of tasks. We're putting it to work in CoCounsel Legal, with more capabilities and sovereign AI options to come. This is the bar we intend to keep raising," Hasker said.