TORONTO, August 24, 2026 – Thomson Reuters (Nasdaq/TSX: TRI), a world contented and exertion company, coming announced the motorboat of Thomson, the company's first proprietary ample connection model, developed in-house. Frontier labs person typically spent billions of dollars connected compute and years of infrastructure finance to scope the frontier. Thomson Reuters took a different path: starting from a beardown open-source instauration and investing $40 cardinal to train Thomson into the correct intelligence for the jobs that matter most, covering talent and compute. The consequence is simply a exemplary Thomson Reuters afloat controls, without the dense conclusion costs of emblematic frontier models.
As 1 of the world's starring providers of trusted contented and expertise for professionals, Thomson Reuters built Thomson connected decades of proprietary content, technology, and domain expertise nary different institution tin match. Training connected that instauration is what made Thomson possible: a exemplary built to Fiduciary-Grade™ standards, astatine a fraction of the emblematic cost.
“For years, the AI manufacture has treated standard arsenic the answer: bigger models, much compute, much money. Thomson shows location is different path,” said Joel Hron, Chief Technology Officer, Thomson Reuters. “Start pinch a beardown foundation, specialize it profoundly for the activity that matters, and you tin build intelligence that is highly capable, acold much businesslike and wholly nether your control. We deliberation that changes the economics of master AI.”
What Makes Thomson Different
Thomson starts from a beardown open-source foundation. What makes it different is what happens next: state-of-the-art mid-training and post-training techniques, drafting connected decades of charismatic contented from Westlaw, Practical Law, Checkpoint, and Reuters, pinch hundreds of taxable matter experts integrated from the creation of training objectives done to the last evaluations.
“Thomson proves what’s imaginable erstwhile you build AI connected decades of proprietary contented and editorial expertise,” said Steve Hasker, CEO of Thomson Reuters. “That’s an advantage only Thomson Reuters has, and it shows successful the results: our early evaluations put Thomson connected par pinch the latest frontier models crossed a scope of tasks. We’re putting it to activity successful CoCounsel Legal, pinch much capabilities and sovereign AI options to come. This is the barroom we intend to support raising.”
The exemplary has been trained connected little than 10% of Thomson Reuters contented truthful far, and what comes adjacent is not simply feeding it much data. It is continued find of caller kinds of specialization and understanding, made imaginable only by building connected decades of proprietary contented and editorial expertise.
AI Sovereignty, and Why It Matters Now
Professionals are paying person attraction to questions of AI sovereignty: really a exemplary is trained, what behaviors and biases unrecorded wrong it, wherever it runs, and really the privateness of their accusation is protected. Thomson marks a displacement for Thomson Reuters into a world wherever those questions are answered directly, not near to 3rd parties alone.
Thomson shows a meaningful uplift from its guidelines exemplary successful instruction following, the expertise to execute complex, multi-part master instructions precisely. It demonstrates an moreover greater uplift successful navigating dense, domain-specific content, the benignant of nuanced reasoning the hardest master tasks require. It is besides capable to beryllium trained alongside Thomson Reuters proprietary devices for illustration Westlaw and Practical Law, which makes it much blase and nuanced successful its work.
The domain-specific summation challenges a communal assumption, that the astir tin general-purpose models only request entree to the correct contented to execute astatine an master level. Thomson Reuters’ early results propose otherwise. Proprietary training and quality taxable matter expertise, applied to a beardown foundation, produces gains that contented entree unsocial does not.
Evaluations of Thomson's underlying instauration exemplary are disposable successful the method study astir the model’s development.
Put To the Test
Ahead of today's launch, Thomson Reuters began opening the exemplary to a group of ineligible and AI academics for nonstop evaluation. We will proceed to make the exemplary disposable to outer parties to assistance successful the further validation and improvement of Thomson complete the coming weeks and months. Thomson Reuters is besides making a “small” type of Thomson disposable arsenic an open-weight exemplary connected Hugging Face for world and non-commercial usage to further assistance successful this validation.
“I tested Thomson against ChatGPT and Claude utilizing immoderate of the much challenging questions students person asked successful my Corporate Tax class. All 3 models answered the questions correctly, but I preferred Thomson's responses overall. I particularly appreciated the links to treatises, which made its responses much transparent and useful for ineligible work.” - Jonathan H. Choi, Washington University School of Law
“Our information recovered Thomson’s citation value mostly competitory pinch starring frontier models, moreover erstwhile tested connected Canadian employment-law questions without a Canada-specific setting.” - Professor Samuel Dahan, Director, Queen's Conflict Analytics Lab and Cornell Legal AI Lab
Trust arsenic the Real Differentiator
Thomson Reuters is processing domain-specific AI for customers pinch the highest expectations of spot and accuracy. The AI manufacture has spent years competing connected earthy capability. Thomson Reuters is betting the adjacent sky will beryllium won successful the verification layer. This supports the early of Fiduciary-Grade AI™ successful practice, the modular Thomson Reuters sets for AI designed for professionals pinch duties of attraction and accountability, wherever almost correct is not bully enough, and customer information is not utilized to train the exemplary without definitive consent.
For CoCounsel, and More
Thomson's first deployment is wrong Tabular Analysis successful CoCounsel Legal, precisely the benignant of high-volume, system archive reappraisal wherever a purpose-built model's advantage shows up immediately. CoCounsel Legal remains multi-model by design, applying Thomson wherever it delivers the clearest advantage and different starring models elsewhere. Thomson will beryllium disposable successful Tabular Analysis for rule firms and firm ineligible departments successful the upcoming release. There are besides plans to widen Thomson models crossed the ineligible and taxation portfolio pinch much sovereign AI options to follow.
The motorboat of Thomson marks a caller section for Thomson Reuters. The institution has ever owned the content, the expertise, and the devices professionals trust connected each day. Now it owns the exemplary too. Thomson Reuters is nary longer only integrating the world's champion content, exertion and expertise. It is building intelligence that will powerfulness the early of master work.
Thomson Reuters
Thomson Reuters (TSX/Nasdaq: TRI) informs the measurement guardant by bringing together the trusted contented and exertion that group and organizations request to make the correct decisions. The institution serves professionals crossed legal, tax, audit, accounting, compliance, government, and media. Its products harvester highly specialized package and insights to empower professionals pinch the data, intelligence, and solutions needed to make informed decisions, and to thief institutions successful their pursuit of justice, truth and transparency. Reuters, portion of Thomson Reuters, is simply a world starring supplier of trusted publicity and news. For much information, sojourn thomsonreuters.com.
Media Contact
Ali Hughes
Director, AI and Innovation Communications
[email protected]
English (US) · 
Indonesian (ID) · 
