Z.ai launches GLM-Legal, a legal AI built on GLM-5.3-Flash, with benchmark claims and a beta for law firms

Z.ai, the developer behind the GLM series, has announced GLM-Legal, a new model built on GLM-5.3-Flash and designed for legal workflows such as contract review, evidence analysis, and compliance. The announcement, posted by the official Zai_org account on Oct. 12, says the model matches frontier models on mainstream legal benchmarks and includes links to a tech blog, API access application, and a tool called AutoClaw. The post also says beta access is open to law firms and legal teams, though no specific launch date was given beyond the posting time. A separate post by the account thehypedotnews reports benchmark results, claiming GLM-Legal ranks first on Harvey's legal agent benchmark across 24 practice areas, significantly outperforming Mistral Large 4 by 21.5% relative to its score, and achieving a score 3x higher than GLM-5.3-Flash on the same benchmark. The same post says GLM-Legal ranks third on LegalBench, trailing Kimi K3 by just 0.02% and slightly behind GPT-6 Astra in second and Claude Fable 5 in first, and second on PlawBench, trailing Claude Fable 5 by 0.04%. Another post by seirdotmk describes a demo the model performed, involving 40 documents from a software project dispute, where the model connected internal warnings, a completion snapshot of about 41%, a client report claiming 60% completion, and a CEO instruction to suppress risk, then built a concealment theory while noting remaining claim elements need testing. That post also reports an honest number of 19.2% all-pass on Harvey LAB, up 12.5 points from the base model, and says most tasks still don't pass strict grading. These reports are from individual posts and have not been independently verified.