Thomson Reuters built its own AI model that now ranks among the best
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Thomson Reuters unveiled "Thomson," a proprietary AI model that benchmarks competitively with Claude Opus 4.8 and outperforms GPT-5.5, Claude Sonnet 5, and Gemini 3.1 Pro across legal and general capability tests.
- Thomson was built on an open-source foundation and trained on decades of proprietary content from Westlaw, Practical Law, Checkpoint, and Reuters news, with outputs evaluated by hundreds of in-house subject matter experts.
- Less than 10% of Thomson Reuters' content has been used in Thomson's training so far, with the company flagging significant remaining room to expand the model's domain knowledge.
- Thomson launches in August as the default model behind Tabular Analysis in CoCounsel Legal, and will roll out across Thomson Reuters' legal and tax product portfolio over the next year.
- On 53 expert-written legal research queries, Thomson—with native access to Westlaw and Practical Law—beat frontier models given unrestricted web access on both completeness and factuality.
- Thomson Reuters acquired Safe Sign Technologies in 2024, betting that professional AI demands domain-specific models rather than off-the-shelf general-purpose intelligence, and pledges customer data is never used to train Thomson.
Why it matters: Thomson Reuters is the first major content-and-tools company to move from AI consumer to AI builder, training on proprietary legal content that general-purpose providers cannot access. On 53 expert legal queries, Thomson beat frontier models with unrestricted web access on completeness and factuality, and the company pledges customer data is never used for training—a key concern for law firms weighing vendor AI tools.
