Thomson Reuters Unveils Proprietary AI Model 'Thomson'

TORONTO, August 24, 2026 – Thomson Reuters, a global leader in content and technology, has announced the debut of its proprietary large language model, named Thomson. Unlike typical frontier models that require immense investment in compute power and infrastructure, Thomson Reuters adopted a different strategy. They leveraged a robust open-source foundation and invested $40 million to train Thomson for critical professional tasks, focusing on specialized intelligence rather than just raw scale. This approach ensures Thomson Reuters has complete control over the model and avoids the high inference costs associated with many advanced AI systems.

As a premier provider of trusted professional content and expertise, Thomson Reuters built Thomson upon decades of proprietary content, cutting-edge technology, and unparalleled domain knowledge. This unique foundation enabled the development of a model that adheres to Fiduciary-Grade™ standards at a significantly reduced cost compared to industry norms.

“For years, the AI industry has focused on scale: bigger models, more compute, more money. Thomson demonstrates an alternative path,” stated Joel Hron, Chief Technology Officer at Thomson Reuters. “By starting with a strong foundation and deeply specializing it for essential work, we've created a highly capable, efficient, and fully controllable intelligence. We believe this fundamentally alters the economics of professional AI.”

Key Differentiators of Thomson

Thomson's distinctiveness lies in its post-open-source development. It employs advanced mid-training and post-training techniques, drawing on decades of authoritative content from platforms like Westlaw, Practical Law, Checkpoint, and Reuters. Hundreds of subject matter experts have been involved throughout the entire process, from defining training objectives to final evaluations.

“Thomson is a testament to what's achievable when AI is built upon decades of proprietary content and editorial expertise,” commented Steve Hasker, CEO of Thomson Reuters. “This is an exclusive advantage for Thomson Reuters, and the results speak for themselves: early evaluations show Thomson performing on par with the latest frontier models across various tasks. We are already integrating it into CoCounsel Legal, with plans for enhanced capabilities and sovereign AI options. We are committed to continuously raising this standard.”

Currently, Thomson has been trained on less than 10% of Thomson Reuters' content. Future development will focus on discovering new specializations and deeper understanding, powered by its extensive proprietary content and editorial expertise, rather than simply increasing data volume.

AI Sovereignty: A Growing Imperative

Professionals are increasingly concerned with AI sovereignty—understanding how models are trained, their inherent behaviors and biases, their operational locations, and the privacy of their data. Thomson marks a significant step for Thomson Reuters in directly addressing these concerns, rather than relying solely on third parties.

Thomson demonstrates a notable improvement in instruction following, enabling precise execution of complex, multi-part professional instructions. It shows even greater proficiency in navigating dense, domain-specific content, which is crucial for demanding professional tasks. Furthermore, its ability to train alongside proprietary Thomson Reuters tools like Westlaw and Practical Law enhances its sophistication and nuanced performance.

The domain-specific gains challenge the prevailing assumption that general-purpose models can achieve expert-level performance simply by accessing the right content. Early results from Thomson Reuters suggest that proprietary training and human subject matter expertise, applied to a solid foundation, yield advantages unattainable through content access alone.

A technical report detailing the development of Thomson's underlying foundation model is available for review.

Rigorous Testing and Validation

Prior to its launch, Thomson Reuters provided early access to the model for evaluation by legal and AI academics. The company plans to continue making the model available to external parties to facilitate further validation and development. Additionally, a smaller version of Thomson is being released as an open-weight model on Hugging Face for academic and non-commercial use.

“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.” - 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.” - Professor Samuel Dahan, Director, Queen's Conflict Analytics Lab and Cornell Legal AI Lab

Trust: The Ultimate Differentiator

Thomson Reuters is developing domain-specific AI for clients who prioritize trust and accuracy above all else. While the AI industry has largely competed on raw capability, Thomson Reuters believes the future lies in the verification layer. This commitment underpins the practical application of Fiduciary-Grade AI™, a standard set by Thomson Reuters for AI designed for professionals with duties of care and accountability, where precision is paramount, and customer data is not used for training without explicit consent.

Applications: CoCounsel and Beyond

Thomson's initial deployment is within Tabular Analysis in CoCounsel Legal, an ideal use case for a purpose-built model in high-volume, structured document review. CoCounsel Legal remains a multi-model platform, utilizing Thomson where it offers the most significant advantage and other leading models elsewhere. Thomson will be available in Tabular Analysis for law firms and corporate legal departments in the next release. Plans are also in motion to integrate Thomson models across the legal and tax portfolios, with more sovereign AI options to follow.

The launch of Thomson signifies a new era for Thomson Reuters. The company has always owned the content, expertise, and tools essential for professionals. Now, it also owns the model. Thomson Reuters is no longer just integrating the world's best content, technology, and expertise; it is building the intelligence that will power the future of professional work.