Z.ai Launches GLM-Legal, Scores 19.2% on Harvey LAB Test

Z.ai's new legal agent scored 19.2% on Harvey's LAB benchmark. The company announced GLM-Legal on October 12 in a post on its own blog, reporting a 12.5 percentage point improvement over its GLM-5.3-Flash base model across the benchmark's 24 practice areas.

Harvey's LAB is strict. The grading is all-pass, meaning "a task counts for the model only if every step lands: the right issues identified, the right provisions cited, the analysis defensible, the output formatted to the brief." The post describes GLM-Legal as a system that "parses specialized contracts, merger filings, and judicial opinions with the precision of top closed-source frontier models."

Under the hood sits a 320 billion parameter mixture-of-experts base, with 18 billion active per token and a 1 million token context window. The company says GLM-Legal was built via "specialized mid-training coupled with high-efficiency post-training," including "reinforcement learning across tens of thousands of realistic legal environments."

Z.ai also claims parity with frontier closed-source systems on LegalBench and the Chinese-focused PLawBench. The post does not name the comparison systems or publish their scores, and no third-party has replicated the Harvey LAB figure.

There is no public API yet. Prospective users must apply via a consultation form, and the model also ships inside AutoClaw, Z.ai's desktop agent, "with $24 of free credits at signup." AutoClaw runs locally on Windows 10 and later, macOS on Apple Silicon and Intel, iOS, and Android, and can operate local files, office software, browsers, messaging tools, and built-in skills from natural-language instructions. The announcement names no Z.ai spokesperson.