Thomson Reuters has launched its first proprietary large language model, allowing it to move away from the high inference costs of typical frontier models.
The company spent $40m to train the model – known as Thomson – using its library of legal content, significantly less than the billions of dollars Thomson Reuters said its rivals typically spend to create the most advanced LLMs.
Working to what it called a "fiduciary-grade" standard, Thomson will initially be used for high-volume document review in CoCounsel Legal, Thomson Reuters' flagship AI legal product. There are plans to extend Thomson models across the company's legal and tax portfolio, with more sovereign AI options to follow.
Joel Hron, Thomson Reuters' chief technology officer, said: “For years, the AI industry has treated scale as the answer: bigger models, more compute, more money. Thomson shows there is another path.”
“Start with a strong foundation, specialise 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.”
Thomson has its origins in Thomson Reuters' 2024 acquisition of UK-based startup Safe Sign Technologies, which specialised in developing legal-specific LLMs. The team now heads foundational legal AI research within Thomson Reuters.
Thomson was built with an open-source foundation and was trained using content from Westlaw, Practical Law, Checkpoint and Reuters, with hundreds of subject matter experts involved from the design stage through to the final evaluations.
The model has been trained on less than 10% of Thomson Reuters content so far. Nevertheless, the company said Thomson showed a "meaningful uplift" from its base model in instruction following and in navigating dense, domain-specific content.
Thomson Reuters is betting that the domain-specific gain can challenge the assumption that the most capable general-purpose models just need access to the right content to perform at an expert level.
"Thomson Reuters’ early results suggest otherwise. Proprietary training and human subject matter expertise, applied to a strong foundation, produces gains that content access alone does not," the company said.
Steve Hasker, Thomson Reuters' CEO, said that Thomson “proves what’s possible when you build AI on decades of proprietary content and editorial expertise.
“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,” he added.
Last week, the company also unveiled the next generation of CoCounsel Legal, which includes an integrated platform for legal research, drafting, legal intelligence and verification.
Email your news and story ideas to: [email protected]
