Z.ai unveils GLM-Legal for legal work, with beta access by application
Built on GLM-5.3-Flash, the specialized model targets legal analysis and drafting. Z.ai reports benchmark gains, while access remains an application-based beta.
A model for legal workflows
Z.ai introduced GLM-Legal on October 12, pitching a specialist model for work that connects source documents, legal reasoning and client-ready writing. The company says it built the model on GLM-5.3-Flash using legal-focused mid-training and post-training. Its English announcement walks through three examples: revising a contract’s exit terms, linking records in a software dispute, and planning privacy compliance across three Southeast Asian markets. Those examples illustrate the intended workflow; they are demonstrations supplied by Z.ai, not independently observed client outcomes.
What the performance claim means
Z.ai reports a 19.2% all-pass rate on Harvey LAB, 12.5 percentage points above its base model, and claims competitive results on LegalBench and PLawBench. The published comparison is the company’s evaluation. It does not establish that the model is reliable across legal jurisdictions or ready to replace professional review. Z.ai is inviting applications for beta API access, while describing its AutoClaw desktop integration as forthcoming. That distinction matters: the announcement introduces a model and a path to test it, not broad public availability or proven performance in live legal matters.