> ## Content Index
> Fetch the complete content index at: https://www.metatalks.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Thomson Reuters built its own legal model on Alibaba's Qwen
- URL: https://www.metatalks.ai/thomson-reuters-launches-thomson-1-0-its-own-legal-model-built-for-about-40-million/
- Published: 2026-08-27T13:07:00.000Z
- Updated: 2026-08-27T13:06:59.000Z
- Author: Al
- Tags: News, AI Model Ownership, Rented Intelligence, #newswire

**Thomson Reuters has released Thomson 1.0, a large language model trained on the legal content it owns and built to take over work it has been renting from Anthropic and OpenAI.**

[Artificial Lawyer reported](https://www.artificiallawyer.com/2026/08/24/tr-launches-thomson-1-0-its-own-llm/?ref=metatalks.ai) that Thomson 1.0 went into production earlier this month. A Thomson Reuters team working with Imperial College London spent months adapting Qwen into a base model the company calls Snowdon, which chief technology officer Joel Hron describes as “ethically and politically de-biased and safe to use.” Chief executive Steve Hasker would not name the open weights when asked; the company's own [engineering post](https://www.thomsonreuters.com/en-us/posts/innovation/how-we-built-thomson/?ref=metatalks.ai) names Qwen.

Hron said the sums charged by labs including Anthropic and OpenAI prompted the build. A model tuned hard for legal work runs on Thomson Reuters' own GPUs and draws on intellectual property the company already holds. It expects that to cost less than renting a frontier model, and says it will pass the difference to customers.

The expertise came first. Thomson Reuters acquired Safe Sign, a Cambridge startup founded by former Google DeepMind scientists, in August 2024, then spent about 20 months and roughly $40 million building the model. Less than 10% of its legal content has gone into it so far.

The deployment starts narrow. Thomson's first job is Tabular Analysis in CoCounsel Legal, the function that reviews documents in bulk, and it reaches law firms and corporate legal teams in an upcoming release. Claude still runs most of CoCounsel — Thomson Reuters broadened its [Anthropic partnership](https://www.thomsonreuters.com/en/press-releases/2026/may/thomson-reuters-and-anthropic-expand-partnership-to-connect-claude-with-cocounsel-legal?ref=metatalks.ai) in May — and work will pass across in stages, only where the company's own expertise brings a benefit. Hasker said Thomson models will spread through the tax and legal products, with sovereign AI options to follow.

The company says early evaluations put Thomson on par with the latest frontier systems across a range of tasks. Those figures are its own, and the published results are mixed rather than a sweep. Independent academic benchmarking is not finished.

Anthropic has accused Chinese labs of illicitly distilling its model outputs and pressed Washington for restrictions, and Senator Tom Cotton has raised security concerns about US companies using Chinese open-weight models. Stanford's James Landay argues that downloadable weights still hide the training data, so a claim that bias was removed is not one a customer can check.

---

Cover image: “blue locker on brown wooden shelf” by [Nasser Eledroos](https://unsplash.com/@nasser%5F?ref=metatalks.ai), [Unsplash](https://unsplash.com/photos/blue-locker-on-brown-wooden-shelf-20fmGtxKs%5FI?ref=metatalks.ai), [Unsplash License](https://unsplash.com/license?ref=metatalks.ai), cropped, resized, re-encoded.