DevOps / SRE / Platform · 24.08.2026, 13:16 UTC
Thomson Reuters trained its own AI model. Then it kept using Anthropic’s anyway.
| Schweregrad | info |
|---|---|
| Kategorie | DevOps / SRE / Platform |
| Quelle | The New Stack ↗ |
| Veröffentlicht | 24.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
Thomson Reuters has developed its own AI model for legal, tax and compliance work, trained on the company’s proprietary professional content and designed to power features inside products such as CoCounsel. Early benchmarks show Thomson competing with models from OpenAI, Anthropic and Google across several professional and general-purpose evaluations.
But Thomson wasn’t built from the ground up. The company started with an existing open-source foundation and spent approximately $40 million training the model, including compute and talent, the company told The New Stack.
“Most of our investment went into further training on decades of proprietary content and expert-driven evaluation, not pre-training a foundation model from scratch,” Thomson Reuters tells The New Stack.
For companies sitting on years of proprietary data, that approach opens another option other than relying entirely on models from OpenAI, Anthropic or Google and spending billions trying to build their own.
Proprietary data as moat
The model was trained using content from Thomson Reuters’ own collection, including Westlaw, Practical Law, Checkpoint, and Reuters. Hundreds of subject-matter experts were involved in evaluating outputs and finding places where the model failed. Thomson Reuters says it has used less than 10% of the content available to it for Thomson’s training so far.
Thomson Reuters still uses frontier models elsewhere in its products. Its new CoCounsel Legal, for example, is built on Anthropic’s Claude Agent SDK. That same SDK is regularly showing up across industries — Spline recently …