GLM-Legal

Spec

Attribute	Value
Developer	Z.ai
Released	not yet
Announced	2026-10-12
Context window	unknown
Pricing	unknown
License	unknown
Availability	AutoClaw access announced as forthcoming
The announcement identifies GLM-5.3-Flash as the base model. It does not establish GLM-Legal's own context window, price, license or downloadable weights; sibling specifications are not inherited here. (source)

Release Date

Announced 2026-10-12. The post says users will soon be able to use the capabilities in AutoClaw; a public availability date is not supplied. (source)

Benchmarks

Z.ai reports a 19.2% all-pass rate on Harvey LAB, covering 24 practice areas, 12.5 percentage points above its base model. It says Vals independently evaluated Harvey LAB and LegalBench, but that evaluation was not separately read. LegalBench and PLawBench parity claims in the announcement lack numerical scores in its static prose. (source)

Use Cases

The examples cover contract review, evidence analysis and cross-border compliance planning. Training combines legal mid-training with RL across tens of thousands of legal environments, co-evolving the model and its execution harness through practitioner feedback. The examples demonstrate intended workflows; they do not establish reliable performance on every legal matter. (source)

Compared To

The base is GLM-5.3-Flash. The reported improvement is tied to Harvey LAB's all-pass criterion; it does not establish a general advantage over every frontier model. The model-harness coupling also connects this release to Agentic Reinforcement Learning and Eval Harness Configuration. (source)

Sources

• Z.ai announcement, 2026-10-12; immutable page-module snapshot.