Z.ai's legal AI model outperforms major competitors in legal benchmark

Z.ai has launched GLM-Legal, a legal AI model designed for contract review, evidence analysis, and compliance research. The model, built on GLM-5.3-Flash, achieved a 19.2% all-pass rate on Harvey’s Legal Agent Benchmark (LAB), nearly three times its base model’s 6.7%. GLM-Legal is a domain-specialized version of the GLM-5.3-Flash, with 320B total parameters and 18B active parameters. It supports text, images, video, and files, and uses sparse and linear attention to reduce long-context serving costs. On the Harvey LAB leaderboard, GLM-Legal ranks #7 out of 79 models with a score of 19.17%, surpassing Kimi K3 (12.92%) and Claude Fable 5 (11.25%). Vals AI confirmed the results, noting a 12.5-point improvement over the base model. Z.ai also reported that GLM-Legal scores 86.23% on LegalBench, trailing Claude Fable 5 at 88.56%. The model’s API is currently application-only, with no pricing details provided. Z.ai plans to integrate GLM-Legal into its desktop agent app, AutoClaw, but has not set a release date.