On October 12, Zhipu announced GLM-Legal, a legal model based on GLM-5.3-Flash and trained on additional legal corpora. It is designed for tasks such as contract review and case-material analysis.
On Vals’ Harvey legal-agent benchmark, GLM-Legal achieved an overall task pass rate of 19.17%, compared with 6.67% for the base model—an increase of 12.5 percentage points. A task counts as passed only if it passes every check. The final score is the average task pass rate given by two AI evaluators.
Access to the API currently requires an application for testing. Integration with the AutoClaw desktop agent is still planned.
On Vals’ Harvey legal-agent benchmark, GLM-Legal achieved an overall task pass rate of 19.17%, compared with 6.67% for the base model—an increase of 12.5 percentage points. A task counts as passed only if it passes every check. The final score is the average task pass rate given by two AI evaluators.
Access to the API currently requires an application for testing. Integration with the AutoClaw desktop agent is still planned.