Thomson Reuters AI Rivals GPT-4, Claude; Benchmarks Revealed

Thomson Reuters states its in-house AI model, called Thomson, now rivals top general-purpose AI models. The company released benchmarking results to support this claim. Thomson performed competitively with Anthropic’s Claude Opus 4.8 and surpassed OpenAI’s GPT-5.5 and Anthropic’s Claude Sonnet 5 in some tests. It also outperformed Google’s Gemini 3.1 Pro in several benchmarks. Thomson is built on an open-source foundation model. It is trained using proprietary content from Westlaw, Practical Law, Checkpoint, and Reuters. Hundreds of subject-matter experts evaluated its outputs. The model's first deployment is set for August, powering Tabular Analysis in CoCounsel Legal. This development shows that highly capable AI models are emerging from more diverse sources.