Skip to content

Back to the catalog

z.aiarticle

GLM-Legal

Z.ai introduces GLM-Legal, a legal-domain model built on the lightweight GLM-5.3-Flash. It is trained with expanded legal-corpus mid-training, then reinforcement learning across tens of thousands of realistic legal environments, with the agent harness and model training iterated together. It reports a 19.2% all-pass rate on Harvey LAB (+12.5 points over the base model) and parity with frontier closed models on LegalBench and PLawBench (Chinese legal practice). Side-by-side examples cover contract review with tracked changes, contradiction mapping across 40 dispute documents, and a three-country privacy compliance plan. Access is via an API application and the AutoClaw desktop agent.

TL;DR Z.ai's GLM-Legal is a GLM-5.3-Flash-based legal model, trained with mid-training plus harness-coupled RL, scoring 19.2% all-pass on Harvey LAB and matching frontier models on LegalBench and PLawBench.
Open original ↗

Organisations named here

  • Z.ai
  • Harvey
  • GLM-5.3-Flash
  • Harvey LAB
  • LegalBench
  • PLawBench

Technologies named here

  • GLM-Legal
  • AutoClaw
  • GLM 5.3 Flash
  • Harness
  • GLM

Techniques named here

  • Specialized mid-training
  • Reinforcement learning across realistic legal environments
  • Harness and model co-iterative optimization
  • Practitioner criteria as engineering loss signals

Related in models