Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LAB
Z.ai introduces GLM-Legal, a legal AI model built on GLM-5.3-Flash, achieving a 19.2% all-pass score on Harvey LAB at $0.15 per test.
Stock photo for illustration only, not from the actual event
- Z.ai has launched GLM-Legal, a specialized legal AI model built upon the GLM-5.3-Flash architecture.
- It achieved a 19.2% all-pass score on Harvey’s Legal Agent Benchmark, nearly tripling its base model performance.
- Testing costs $0.15 per run, with access currently restricted to application-only submissions.
Z.ai has made waves in the artificial intelligence sector by unveiling GLM-Legal, a dedicated legal AI model powered by the GLM-5.3-Flash foundation. Engineered to tackle intricate legal workflows, this model has demonstrated remarkable performance on industry benchmarks, successfully outperforming competitors such as Claude Fable 5 and Kimi K3 on Harvey LAB.
According to benchmark evaluations on Harvey LAB, GLM-Legal secured a 19.2% all-pass score. This figure highlights a substantial performance leap, representing nearly triple the capability of its base model counterpart and underscoring significant advancements in legal comprehension and reasoning accuracy.
Regarding cost-efficiency and accessibility, Z.ai prices the testing of the GLM-Legal model at just $0.15 per test. Despite its economic testing structure, access to the platform is not yet publicly open. Interested users must go through an application process in order to gain entry and evaluate the system.
Stock photo for illustration only, not from the actual event
The rise of domain-specific models like GLM-Legal outperforming general benchmarks highlights a broader industry shift toward specialized artificial intelligence. In the legal sector, such precise tools promise to dramatically reduce the time and cost required for document review and contract analysis, signaling a new era of legal tech adoption.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment