CoCounsel Legal

CoCounsel Legal is an AI work assistant for legal professionals. When lawyers or legal teams review contracts, look up case law, or draft documents, it acts like a junior researcher sitting beside them, digging up materials and writing first drafts. It was built by Thomson Reuters, the legal and financial information company, and draws on the decades of case-law databases and legal literature the company has built up as raw material for its answers.

The engine underneath this service, the language model it runs on, has changed over time. At first it simply borrowed a general-purpose language model built by another company. But recently, Thomson Reuters unveiled a model of its own called Thomson, built by taking an open-source model released by Alibaba and retraining it in-house. It's as if a carmaker that used to buy engines from someone else built its own engine factory.

Why bother building its own engine? The company's explanation is that for high-volume, repetitive work of the same type, like reviewing contracts, running your own model turns out to be cheaper than paying a monthly subscription to rent someone else's.

The name CoCounsel doesn't actually appear in the article. Instead, it shows up in a sentence like 'Thomson Reuters unveiled its own language model, Thomson' — and this Thomson model is the engine that runs Thomson Reuters legal AI services like CoCounsel Legal behind the scenes. The tricky part is the phrase 'built in-house,' which doesn't mean it was made entirely from scratch. In reality, it was created by taking an open-source model released by Alibaba and retraining it on Thomson Reuters' legal data.