Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LAB
GLM-Legal is a legal AI model from Z.ai, built on GLM-5.3-Flash and trained for agentic legal work.
Proof2 independent outlets
Key points
- Z.ai lists no separate GLM-Legal parameter count.
- Best: 19.17% Harvey LAB task pass rate, #7 of 79 on Vals AI, at $0.15 per test.
- GLM-Legal is a domain-specialized version of GLM-5.3-Flash for law firms, in-house counsel and compliance teams.
- Z.ai’s legal AI model for contract review and compliance
Sources (2)
- [1]Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LABMarkTechPost · Oct 13, 08:19 AM
GLM-Legal is a legal AI model from Z.ai, built on GLM-5.3-Flash and trained for agentic legal work.
Z.ai lists no separate GLM-Legal parameter count.
- [2]GLM-LegalProduct Hunt: AI launches · Oct 12, 04:24 PM
Z.ai’s legal AI model for contract review and compliance
Extractive summary: sentences quoted from the sources.
Hype check
Independently evaluated3 independent looks since
- Evaluation · Oct 12, 2026Qwen3.8-27B vs Frontier vs Others
- Evaluation · Oct 12, 2026PhysCodeGym: Benchmarking Coding Agents for Physically Verifiable 4D Reconstruction
- Evaluation · Oct 12, 2026Position: Beyond Training Time, Test-Time Coordination is Essential for Cooperative MARL
- Follow-up · Oct 12, 2026Profiling web-edit with Python 3.15: flame graphs and source heatmaps
- Follow-up · Oct 12, 2026[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
- Follow-up · Oct 12, 2026In 2030 owning a gpu cluster will be seen the same as owning a gun for home defense
Before this
- Oct 12, 2026Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents
- Oct 12, 2026Redesigning source control at Anthropic
- Oct 12, 2026Strengthening our defenses against distillation attacks
- Oct 11, 2026"Strata" for GLM5.3 Flash is here for some! Project Maya
- Oct 11, 2026vllm-ascend updates (GLM5.3-flash) on dual 310p Ascend cards
- Oct 11, 2026A Sovereign, Indic-First, Reasoning Model
