Thomson Reuters Builds $40M Legal AI on Top of Qwen
0xsachi · x · 2026-08-31
Thomson Reuters has launched "Thomson," an in-house LLM built on top of Qwen3.5-397B and fine-tuned on 175 years of proprietary legal, tax, and news data. The project cost $40 million and was developed by the team from Safe Sign, a legal AI startup acquired in 2022.
Key Details:
- Strategy: Leveraging open-source models with proprietary data (using less than 10% of their data collection so far).
- Performance Claims: Internal benchmarks place it comparable to Claude Opus 4.8 and ahead of GPT-5.5 and Claude Sonnet 5 (note: version numbers may be hyperbolic).
- Trend: Signals a shift where data-rich companies build their own models rather than relying solely on general-purpose APIs.
Related event: Thomson Reuters Unveils $40M Legal LLM(2 posts)→
More from Companies & People
- The 'sci-fi make it till you make it' playbook: Musk pioneered it, Altman copied it — Linahuaa · 2026-10-16
- Ex-DeepMind researcher Michal Valko surfaces with stealth AI startup on stage — misovalko · 2026-10-16
- Frontier lab post-training teams draw few grads from top NLP programs — lateinteraction · 2026-10-16
- First world-building hackathon Tripothon SF wraps with World Labs and Runway — mhdfaran · 2026-10-16
- Name the 3 numbers that justify your AI spend — or you're reporting intention — YvesMulkers · 2026-10-16
- Andrew Chen's Lecture 3: You cannot fix bad retention — ideas before tools — richardreis · 2026-10-16