AIEvals

Models

GLM-Legal

Z.ai’s legal specialist based on GLM-5.3-Flash, trained for legal reasoning, document review and professional agent workflows.

Z.aiReleased 12 October 2026Context Not publishedOutput Not publishedPrivate weightsList price not published

Family GLM Legal3 benchmarks with a published resultRead 17 October 2026

Where it stands

Rank compares tracked models within each publisher table. Published includes all reported entries in that table. Independent scores and ranks are not available.

BenchmarkScoreRankPublishedLeaderBehind
PLawBenchZ.ai · publisher table · 12 October 202677.30%2of 42of 5Claude Fable 50.40 pts
LegalBenchZ.ai · publisher table · 12 October 202686.20%3of 43of 5Claude Fable 52.40 pts
Harvey's Legal Agent BenchmarkZ.ai · publisher table · 12 October 202619.20%1of 41of 5GLM-Legal—

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
PLawBenchProfessional knowledgeno published result
77.3%PLawBenchZ.ai · 12 October 2026Rank 2 of 5 in this publisher table
LegalBenchProfessional knowledgeno published result
86.2%LegalBenchZ.ai · 12 October 2026Rank 3 of 5 in this publisher table
Harvey's Legal Agent BenchmarkProfessional knowledgeno published result
19.2%Harvey's Legal Agent BenchmarkZ.ai · 12 October 2026Rank 1 of 5 in this publisher table
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Source notes

From the linked sources. Read 13 October 2026.

Announcement and API access. Announced 12 October 2026 in beta. Apply for API access through Z.ai’s consultation page; the announcement says AutoClaw integration is coming soon. This date records the announcement, not general API availability.

Base model and training. Built on GLM-5.3-Flash with legal mid-training and reinforcement-learning post-training alongside an agent harness. GLM-Legal and its base model retain distinct comparison columns.

Specifications and weights. The announcement does not publish context or output limits, parameter counts, API model identifiers, token prices or a public weight release. The catalogue records private weights for the announced beta; it does not infer specifications from the base model.

Legal evaluations. Z.ai reports Harvey LAB’s all-pass rate across 24 practice areas, plus LegalBench and PLawBench. All 15 comparison figures remain publisher reports. The announcement does not specify reasoning effort, trial count, LegalBench aggregation or the PLawBench judge and task subset.

Sources

Read on 17 October 2026.