Thomson Reuters

Overview

Thomson Reuters (TR) is a Toronto-based professional-information conglomerate best known for Westlaw, Practical Law, Checkpoint, and Reuters — corpora built over decades of legal, tax, and news publishing. In 2025 TR acquired Casetext and integrated its CoCounsel product as the flagship AI-assistant surface across the legal-tech stack. TR has since positioned itself as a data-holder-as-lab: monetising proprietary corpora via post-trained models delivered inside its own products rather than licensing raw data to third-party frontier labs.

Timeline

• 2026-08-25-AI-Digest — Launched “Thomson,” a post-trained frontier-class model built on an open-source foundation at a ~$40M training cost, trained on less than 10% of TR’s proprietary content (Westlaw, Practical Law, Checkpoint, Reuters). Deployed in CoCounsel’s Tabular Analysis feature; small open-weight variant released to Hugging Face. Extends the BloombergGPT (Mar 2023) “domain-incumbent builds in-house” template three years later at seven-figure cost.

Key Developments

1. Post-trained frontier-class model on an open-source base: The “Thomson” model is a $40M fine-tune rather than a from-scratch pretraining run, trained on less than 10% of TR’s proprietary corpus. TR is not disclosing the full data moat; the model is a demonstration of what that moat could produce at scale rather than the definitive product.
2. Open-weight distribution play: TR released a small open-weight variant of the Thomson model to Hugging Face — a data-holder giving away a slice of its post-training work as distribution, rather than keeping everything behind CoCounsel’s paywall.
3. CoCounsel deployment surface: The Thomson model powers CoCounsel’s Tabular Analysis feature; CoCounsel is the AI-assistant product TR acquired via Casetext in 2025 and has since positioned as the primary integration point for AI across Westlaw and Practical Law.