Thomson Reuters is launching its own language model, Thomson, built on Alibaba's Qwen. According to The Decoder, the company plans to spend approximately 40illion on the project over two years.
The source description notes that the model shows its best results when accessing Thomson Reuters' own content, including Westlaw. Chief Technology Officer Joel Hron links the value not only to general intellectual power but also to understanding which capabilities are important for the company to control internally.
The practical implication of this decision is shifting part of the dependence from external model providers to an in-house system and proprietary data. However, available materials do not disclose the testing methodology, comparison conditions, or the actual scale of Thomson's usage.
editorial commentary
Why it matters
If the model proves particularly effective on specialized data, the strategy could strengthen Thomson Reuters' control over legal AI services and reduce dependence on external vendors. The nearest observable signal will be published test results outside of proprietary content. The main uncertainty is the lack of available evaluation methodology and confirmed data on the model's performance in real products.