Pull to refresh
Logo
Thomson Reuters launches Thomson, its own legal AI model

Thomson Reuters launches Thomson, its own legal AI model

New Capabilities

A $40M in-house large language model trained on Westlaw and Practical Law content, built to cut reliance on Anthropic and OpenAI

Today: News coverage consolidates around Thomson's significance

Overview

Updated 1 hour ago

Thomson Reuters on Monday launched Thomson, its first proprietary large language model. The company spent $40 million training it on decades of its own content, including Westlaw, Practical Law, Checkpoint, and Reuters material, starting from an open-source Qwen 3.5 base rather than building from scratch.

The model's first job is inside Tabular Analysis in CoCounsel Legal, a document review feature that processes high-volume structured legal documents. Thomson Reuters says the model is fully owned and controlled by the company, which gives it leverage over deployment, governance, and costs. A smaller version was released as an open-weight model on Hugging Face for academic and non-commercial evaluation.

Why it matters

If Thomson holds up, legal AI no longer needs to rent models from Anthropic or OpenAI — and Westlaw's data becomes a moat, not a feature.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

$40M
Development investment
Total spent on talent and compute to train Thomson.
$450K
Final training run cost
Cost of the last training run that produced the released model.
Qwen 3.5
Open-source base model
Thomson was built on Qwen 3.5, an open model from Alibaba's Ant Group subsidiary.
1
Deployed features at launch
Thomson powers Tabular Analysis in CoCounsel Legal; expansion to more tasks planned.

Voices

Curated perspectives — historical figures and your fellow readers.

Ever wondered what historical figures would say about today's headlines?

Sign up to generate historical perspectives on this story.

Play

Exploring all sides of a story is often best achieved with Play.

Most of these play right now — no account needed. Sign up to save scores, keep a streak, and unlock Debate and Predict. Log in Sign Up
Predict 3 ways this could play out. Back the one you believe — contrarian picks score more when a scenario has a resolution date. Log in to play

People Involved

Organizations Involved

Timeline

July 2026 August 2026

4 events Latest: Today
Tap a bar to jump to that date
  1. News coverage consolidates around Thomson's significance

    Today Media Analysis

    Analysts note the $40M cost, the Qwen 3.5 foundation, and the strategic shift away from renting Anthropic and OpenAI models.

  2. Thomson formally launched

    Product Launch

    Thomson Reuters announced its proprietary LLM, trained on Westlaw and Practical Law content. $40M invested; deployed in CoCounsel Tabular Analysis.

  3. New CoCounsel Legal released for general availability

    Product Release

    The agentic CoCounsel platform, built on Anthropic Claude Agent SDK, added Westlaw Brief Builder and Workspaces.

  4. Preliminary benchmarks released

    Announcement

    Thomson Reuters shared initial benchmarking data for Thomson, showing performance above some general models.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

January 2023

Microsoft's $13B OpenAI partnership (2023)

Microsoft invested roughly $13 billion in OpenAI, securing preferred access to its frontier models and making Azure the exclusive cloud provider. Sam Altman's brief ouster in November 2023 showed how exposed Microsoft was to a partner it didn't control.

Then

Microsoft integrated GPT models across Office, Bing, and Azure, cementing OpenAI as the default enterprise AI provider.

Now

The episode pushed large enterprises to hedge their AI dependence, accelerating interest in open-weight models and multi-provider strategies.

Why this matters now

Thomson Reuters is acting on the same lesson Microsoft learned: renting frontier AI from a partner means surrendering control over cost, governance, and roadmap.

March 2023

BloombergGPT (March 2023)

Bloomberg trained a 50-billion-parameter model on its proprietary financial data and public text, betting that domain-specific training on exclusive data could beat general-purpose models for finance tasks.

Then

BloombergGPT showed strong results on financial benchmarks but never became a dominant product; Bloomberg continued using multiple AI providers.

Now

It became the template for companies with proprietary data building their own domain models rather than paying for general ones.

Why this matters now

Thomson Reuters follows the BloombergGPT playbook but goes further: training on the same data that makes Westlaw and Practical Law industry standards, then deploying it inside an existing AI-powered product.

Sources

(9)