發生什麼事
GLM-Legal is built on GLM-5.3-Flash and is meant for contract review, evidence analysis, compliance, and similar work (@Zai_org). Z.ai said it reaches a 19.2% all-pass rate on the Harvey LAB benchmark, 12.5 percentage points above the base model (Z.ai blog). Third-party evaluator Vals AI records 19.17%, ranking it 7th of 79 models, and lists the weights as open; Z.ai itself only offers API access by application (Vals AI). A cross-border seller account found the official Southeast Asia privacy checklist example the most useful (@harrison_0x821); the same account also asked whether it holds up on non-US documents (@harrison_0x821). Another account wondered how it handles badly formatted contracts (@ElectronDev).
01Sources
- @Zai_org: official GLM-Legal announcement, with tech blog and API application
- Z.ai: GLM-Legal: Advancing Agentic Legal Intelligence: official tech blog, Harvey LAB 19.2%
- Vals AI: GLM Legal: third-party evaluation, 19.17%, 7th of 79 models
- @harrison_0x821: asks about performance on non-US documents
- @harrison_0x821: cross-border seller likes the privacy checklist example
- @ElectronDev: wonders how it handles badly formatted contracts