New Models AI Minute Newsroom 2026-08-25

The final training run cost about $450,000. The company that owns Westlaw now has its own frontier model.

Thomson Reuters announced Thomson on 24 August, the first large language model it has built in house. It was trained on decades of Westlaw, Practical Law, Checkpoint and Reuters content — less than a tenth of the company's archive so far — with hundreds of its own subject-matter experts involved. Total investment in talent and compute came to about $40 million, and the company says the final training run itself cost roughly $450,000. Early evaluations, it says, put Thomson on par with the latest frontier models across a range of tasks, with particular strength in instruction-following and citation quality. It ships first inside Tabular Analysis in CoCounsel Legal, which stays a multi-model product, and a smaller open-weight version is on Hugging Face for academic and non-commercial use.

Why it mattersThe claim that only companies with billions can train a competitive model is being tested from a direction nobody was watching. Thomson Reuters has no compute advantage; what it has is a licensed, curated legal and tax archive that no lab can legally scrape. If a $450,000 run on that archive lands anywhere near frontier quality on the narrow work its customers pay for, then every bank, hospital system and publisher sitting on locked data has an option other than renting someone else's model — and the labs start losing the customers whose text is worth the most.

✓ Verified · 3 sources

Read in the app — free, in 9 languages

The final training run cost about $450,000. The company that owns Westlaw now has its own frontier model.

Related stories