Z.ai introduced GLM-Legal, a legal model based on GLM-5.3-Flash, and invited legal teams and others to join its beta API. It says the model and tool-execution harness were trained together for contract revisions, linking contradictions in evidence, and drafting legal documents. The introductory material reports Harvey LAB’s overall pass rate rising from 6.7% for the base model to 19.2%, while the directly checked Vals AI evaluation records 19.17%. Terms for a generally available public API were not confirmed. The results do not establish accuracy in Korean legal practice or the replacement of professionals.