Thomson Reuters shifts heavy legal-AI tasks to its own model to curb costs

Thomson Reuters says it has moved token-heavy work, including bulk document review and tabular analysis, from frontier models to its own model in CoCounsel. CEO Stephen Hasker says the change has improved speed and accuracy and could help the company keep all-you-can-eat pricing for many customers, rather than passing every extra token on to the bill.

Watch Desk analysis

What happened

Hasker described the change in an interview reported by Diginomica. He said Thomson Reuters had ported the tasks over the previous month and was targeting near-full capacity use on its model to gain a cost advantage. The company says it plans to make use of that advantage over the next year; Hasker called the work early-stage.

The same account says CoCounsel NextGen had, in September, surpassed the previous version’s best month on every engagement and retention measure Thomson Reuters tracks. Its active user base was up more than 60% from August, and users were completing 2.4 times as much work as at the previous version’s peak. Those are company-reported figures, not independent measurements.

Why it matters

This is a practical answer to a growing problem for AI services: heavy use can make the economics awkward, especially when customers expect a predictable subscription. Routing suitable work to a company’s own model may help Thomson Reuters control costs without turning every document review into a metered event.

It also makes the choice of model part of the product’s pricing strategy, not just a technical detail. If the company’s reported speed and accuracy gains hold up, customers could get more work done without paying by the token. The catch is that the claims come from the company whose pricing and product are at stake.

Our read

The interesting move is not simply “use a smaller or cheaper model”. It is shifting a defined, demanding workload and trying to preserve an all-you-can-eat offer. That is a credible test of whether AI providers can absorb heavy usage without making the customer’s bill a thrilling monthly surprise. Buyers should ask how the savings and accuracy are measured, and whether the fixed-price terms apply to their workloads.

What to watch

• Whether Thomson Reuters reports measured cost, speed and accuracy comparisons for the migrated tasks.
• How broadly all-you-can-eat pricing applies, particularly for heavy users.
• Whether the reported NextGen engagement gains translate into sustained use and customer value.

Discussion spark: Should legal AI providers absorb the cost of heavy use through flat pricing, or should customers pay more when they use more?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.