Thomson Reuters has launched Thomson, its first proprietary large language model, giving the company greater control over the technology powering its professional AI products.
Developed in-house and funded with an investment of $40 million, Thomson begins with an open-source foundation that Thomson Reuters further trained using its proprietary content and domain expertise.
That process drew on authoritative material from Westlaw, Practical Law, Checkpoint and Reuters, with hundreds of subject matter experts involved in designing training objectives and evaluating the results.
Thomson Reuters says this approach allowed it to build a specialized model at a fraction of the cost of comparable frontier models while avoiding the high inference costs typically associated with operating them.
“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 at 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.”
The launch also gives Thomson Reuters control over how the model is trained, where it operates and how customer information is protected. The company is positioning this sovereignty as an important advantage for lawyers, accountants and other professionals working with sensitive information.
Thomson Reuters says customer data will not be used to train the model without explicit consent, aligning Thomson with the company’s emerging “Fiduciary-Grade AI” standard for professionals with duties of care and accountability.
According to the company, early evaluations found Thomson performed on par with the latest frontier models across a range of tasks. The model showed particular improvement over its underlying foundation model when following complex instructions and navigating dense, domain-specific content.
Thomson has so far been trained on less than 10 per cent of Thomson Reuters’ available content, leaving room for further specialization across the company’s legal, tax and professional information products.
“Thomson proves what’s possible when you build AI on decades of proprietary content and editorial expertise,” said Thomson Reuters CEO Steve Hasker.
“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.”
Legal and AI academics have also begun evaluating the model. Professor Samuel Dahan, Director of Queen’s Conflict Analytics Lab and Cornell Legal AI Lab, said Thomson’s citation quality was generally competitive with leading frontier models, including when tested on Canadian employment law questions without a Canada-specific setting.
Thomson Reuters plans to continue making the model available to external evaluators. A smaller open-weight version will also be released through Hugging Face for academic and non-commercial use.
The model’s first commercial deployment will be within Tabular Analysis in CoCounsel Legal, where it will support high-volume, structured document review. Thomson Reuters said the model will become available to law firms and corporate legal departments in an upcoming release.
CoCounsel Legal will remain a multi-model platform, using Thomson for tasks where the proprietary model offers the clearest advantage while continuing to use other leading models elsewhere.
Thomson Reuters also plans to extend Thomson models across its legal and tax portfolios, with additional sovereign AI options expected to follow.