Thomson Reuters builds in-house AI model on Alibaba's Qwen for $40M
schwarzjn_ · x · 2026-08-28
Thomson Reuters spent two years and approximately $40 million building an in-house AI model, with the final training run costing just $450,000. The project started with Alibaba's open-source Qwen model and was fine-tuned on proprietary legal and news data from Westlaw and Reuters. The CTO compared relying on external AI APIs to being a permanent tenant, while owning the model equates to owning the building. Enterprise customers are broadly cutting spending on OpenAI and Anthropic in favor of cheaper alternatives.
More from Companies & People
- Revenge IPO goal: Turning grudges into chips in pockets — cixliv · 2026-08-28
- Missing OpenAI or Anthropic is a career-ending move for VCs — beffjezos · 2026-08-28
- Building agents is easy now; safely operating 50 at once is the real bottleneck — Many_Audience7660 · 2026-08-28
- X's link UI update changes the game: how the split view affects engagement — gefei55 · 2026-08-28
- Ex-Nvidia engineer on company culture and inference infrastructure — Scobleizer · 2026-08-28
- Crypto designer Mochi Kuan joins OpenAI's design team after four years at Coinbase — i_dg23 · 2026-08-28