Z.ai · 2026-10-12 · major
GLM-Legal — Z.ai's agentic model for contracts, evidence and compliance
GLM-Legal is Z.ai's legal-domain agent model, built on GLM-5.3-Flash and trained with reinforcement learning on tens of thousands of legal environments. It is in a gated beta for law firms, legal tech and in-house teams.
Z.ai tuned its GLM-5.3-Flash model into an agent that reviews contracts, sifts evidence and researches compliance.
Quick facts
| Maker | Z.ai (formerly Zhipu AI) |
|---|---|
| Base model | GLM-5.3-Flash — 320B total, 18B active |
| Context window | 1M tokens (base model) |
| Harvey LAB | 19.2% all-pass, +12.5 points over the base |
| Access | Gated beta via application form; AutoClaw desktop agent |
What is it?
GLM-Legal is a legal-domain agent model that Z.ai announced on 12 October 2026. Z.ai built it on GLM-5.3-Flash, its 320B-parameter mixture-of-experts model with 18B active parameters, and targets three jobs: contract review, evidence analysis and compliance research across jurisdictions.
How does it work?
Reinforcement learning is the core of the training: Z.ai trained the model across tens of thousands of legal environments rather than only fine-tuning on legal text. Unite.AI reports a 19.2% all-pass rate on Harvey's Legal Agent Benchmark across 24 practice areas, 12.5 points above the GLM-5.3-Flash base, and Z.ai claims results comparable to frontier systems on LegalBench and PLawBench.
Why does it matter?
Legal work is where general models still fail on long, multi-document tasks, and the jump on Harvey LAB shows how much domain RL adds on top of an open base model. Access is limited for now: API use goes through a design-partner application, while the AutoClaw desktop agent offers a way to try Z.ai models with free signup credits.
Who is it for?
law firms, legal tech builders, in-house legal teams
Frequently asked questions
- Can anyone use GLM-Legal today?
- No. GLM-Legal launched as a gated beta on 12 October 2026. Law firms, legal tech companies, in-house legal teams and independent practitioners apply through a Z.ai consultation form that asks about their intended use, and AlphaSignal describes the program as a design-partner arrangement rather than an open API.
- How does GLM-Legal compare to the GLM-5.3-Flash base model?
- On Harvey's Legal Agent Benchmark, GLM-Legal reaches a 19.2% all-pass rate across 24 practice areas, according to Unite.AI, which is 12.5 points above GLM-5.3-Flash. Vals AI scores the base GLM-5.3-Flash at 6.7% on the same benchmark, so the legal training roughly triples the share of fully solved tasks.
- Is GLM-Legal open weights?
- Z.ai has not released GLM-Legal weights. The GLM-5.3-Flash base model is published under an MIT license, but AlphaSignal notes Z.ai did not say whether GLM-Legal would carry the same terms. For now GLM-Legal is reachable only through the gated beta and Z.ai's own products.
- What legal tasks does GLM-Legal handle?
- Z.ai shows GLM-Legal on three kinds of work: reviewing contracts for changed terms such as notice periods, analysing a set of 40 evidence documents to find where a client report contradicts internal records, and researching compliance rules across Singapore, Malaysia and Indonesia.
Try it
Apply for access at z.ai/consultation/legal