Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LAB
Z.ai's legal model hits 19.2% on Harvey's benchmark—nearly 3x its base model. Here's what that means for specialized vertical AI.
Why it matters
Z.ai GLM-Legal demonstrates measurable capability gains on a domain-specific benchmark (Harvey LAB), establishing a data point in the emerging pattern of vertical model specialization. The 3x improvement over base and competitive positioning against Claude and Kimi matters to practitioners evaluating legal AI, but the real story is whether specialized legal models are becoming a category.
The key facts
12 to knowZ.ai GLM-Legal: 19.2% all-pass on Harvey Legal Agent Benchmark
3x improvement over base model (GLM-5.3-Flash implied ~6-7%)
Scoring $0.15 per test (inference pricing disclosed)
Outperforms Claude Fable 5 and Kimi K3 on Harvey LAB (specific competitors named, no margin disclosed)
Access by application only (gated availability)
Built on GLM-5.3-Flash (Alibaba foundation model, not proprietary)
GLM-Legal: 19.2% all-pass on Harvey Legal Agent Benchmark
Base GLM-5.3-Flash: ~6.4% pass rate (implied from 3x improvement claim)
Outperforms Claude Fable 5 and Kimi K3 on Harvey LAB
Cost: $0.15 per test
Access: application-only (not public)
Model: built on GLM-5.3-Flash architecture
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Z.ai’s GLM-Legal is a legal AI model built on GLM-5.3-Flash. It scores 19.2% all-pass on Harvey’s Legal Agent Benchmark, nearly 3x its base, at $0.15 per test. Access is by application.