CONFIRMED (Z.ai blog, 12 Oct 2026). All benchmark figures are COMPANY CLAIMS by Z.ai and have not been independently verified.
Z.ai has introduced GLM-Legal, a model trained to carry out legal work from start to finish: reading source documents, applying the law and producing finished work product. The company built it on “the lightweight GLM-5.3-Flash foundation” [1].
An agentic model doesn’t just answer one question. It works through a task in steps, using tools and documents, until it produces a finished result.
What Z.ai says about performance (COMPANY CLAIM)
- “GLM-Legal achieved a 19.2% all-pass rate on Harvey LAB” [1]. Z.ai says the test covers “real-world law firm workflows across 24 practice areas”, and that GLM-Legal “improves on its base model by 12.5 percentage points” under “Harvey’s strict all-pass grading” [1].
- Z.ai says the model is “matching frontier closed-source models on LegalBench and PLawBench despite running on a significantly leaner parameter footprint” [1]. PLawBench covers Chinese legal consultation, case analysis and document drafting.
All-pass grading means a task counts only if every requirement is met, so even a low-looking percentage can be a hard bar. Z.ai did not say how many parameters the model has.
How it was trained
Z.ai describes two stages [1]:
- Mid-training: “We expanded the volume of high-quality legal corpora, allowing the model to learn legal concepts, statutory rules, and interpretive frameworks within dense domain context.”
- Post-training: reinforcement learning “across tens of thousands of realistic legal environments”. Here the model is trained together with a Harness, the software that “manages task context, connects toolsets, and routes execution feedback”. Z.ai says “Harness architecture and model training iterate in lockstep.”
Reinforcement learning trains a model by rewarding good results on practice tasks, rather than only showing it example text.
Coming to AutoClaw
“You’ll soon be able to put these capabilities to work in AutoClaw, our desktop-native agent platform,” Z.ai says [1]. Its examples include contract review, where “AutoClaw puts those edits into the Word document with tracked changes and comments”, evidence analysis across case documents, and privacy-compliance planning. Z.ai gave no launch date or pricing.
Links: Z.ai legal sign-up · AutoClaw
Sources
- Z.ai, “GLM-Legal: Advancing Agentic Legal Intelligence”, 12 October 2026. https://z.ai/blog/glm-legal
