Thomson Reuters says its in-house large language model, Thomson, now performs competitively with top general-purpose AI models, releasing its first benchmarking results on July 31 via Chief Technology Officer Joel Hron and Head of AI Research Jonathan Schwarz. Developed since the 2024 acquisition of Safe Sign Technologies, Thomson performed competitively with Anthropic’s Claude Opus 4.8 and ahead of OpenAI’s GPT-5.5, Anthropic’s Claude Sonnet 5, and Google’s Gemini 3.1 Pro across legal and general-purpose benchmarks. Starting from an open-source foundation model, Thomson is trained on proprietary content from Westlaw, Practical Law, Checkpoint and Reuters, with less than 10% of that content used so far. Its first deployment comes in August as the default model for Tabular Analysis in CoCounsel Legal, with integration across legal and tax products planned over the next year. In benchmarks, Thomson-1-Large took the top score in three of seven rows, including PrBench Legal Hard with 0.352, ahead of GPT-5.5 (0.333), Opus 4.8 (0.315), and Gemini (0.293). On Stanford LegalBench, Thomson scored 0.823, behind Gemini 3.1 Pro (0.843) and GPT-5.5 (0.832), but ahead of Opus 4.8 (0.818).