🕒 Created

Thomson Reuters launches Thomson 1.0, a proprietary AI model designed to specialize in legal, tax, and compliance work.

Thomson Reuters Corp. has officially launched Thomson 1.0, its first proprietary large language model. The company developed the model to provide a specialized edge in legal, tax, and compliance work, leveraging its decades of proprietary content from sources like Westlaw and Practical Law. Unlike general-purpose models, Thomson 1.0 is designed to set the frontier of intelligence for legal professionals, focusing on high-volume document review and structured analysis. To develop the model, Thomson Reuters spent approximately $40 million over two years, utilizing an open-source foundation and realigning it with the company's values and professional expertise. The final training run cost roughly $450,000. The model is being integrated into the CoCounsel Legal AI assistant, specifically powering the Tabular Analysis feature. Internal testing shows that Thomson 1.0 is competitive with leading frontier models from OpenAI, Anthropic, and Google, particularly when connected to Thomson Reuters' authoritative content. The company aims to provide a more efficient and controlled environment where the model is trained to flag uncertainty rather than force confident-sounding answers, ensuring higher accuracy for professional use.

Sources