0
Applied AI·August 24, 2026·1 min read

Thomson Reuters built its own model on a Chinese open base

Share

Spending $40M to train an in-house LLM on top of a Chinese open-source base is a bet that domain control and IP certainty matter more than chasing frontier benchmarks. If you're a data-rich incumbent, the question this week is whether your proprietary corpus justifies a similar vertical model layer rather than living entirely on general-purpose APIs.