header-langage
简体中文
繁體中文
English
Tiếng Việt
한국어
日本語
ภาษาไทย
Türkçe
Scan to Download the APP
Partnered With:

Reuters' parent company finds Claude too expensive: spends $40 million to build a large model itself, with the base from Qwen

Exclusive Report from Beating AI Newsletter: Thomson Reuters Launches First Self-developed Large Model Thomson, Ready to Gradually Take Over Some of Claude's Current Tasks. The company has invested approximately $40 million in total, mainly focusing on talent and computing power.

Thomson Reuters CTO Joel Hron confirmed to Business Insider that Thomson is based on Snowdon, which is a model developed by a team from Imperial College London using Ali Qwen. Thomson Reuters further trained the model using in-house data such as Westlaw, Practical Law, Checkpoint, and Reuters, with the current content utilization being less than 10% of all the data.

Hron likened using external models to "renting a house": you keep paying someone else, and you don't control the model's price and roadmap. Thomson Reuters' legal AI assistant, CoCounsel, still primarily uses Claude, but the company plans for Thomson to take over more and more functions, starting with high-frequency tasks like large-scale document review.

Correction/Report
Submit
Add Library
Visible to myself only
Public
Save
Choose Library
Add Library
Cancel
Finish