Z.ai GLM-Legal Tops Harvey LAB Beating Claude & Kimi

Z.ai has officially launched GLM-Legal, a specialized legal reasoning model that achieves unprecedented performance on Harvey's Legal Agent Benchmark (Harvey LAB). Officially released on October 13, 2026, the model scores 19.2% all-pass accuracy, nearly tripling the performance of its base model while outperforming established competitors including Claude Fable 5 and Kimi K3. The launch positions GLM-Legal as a cost-effective solution for legal professionals requiring advanced AI reasoning capabilities.

Performance Benchmarks and Competitive Edge

GLM-Legal demonstrates a significant leap in legal reasoning performance with its 19.2% all-pass score on Harvey LAB, the industry-standard benchmark for evaluating legal AI systems. This achievement represents nearly three times the performance of its foundation model, GLM-5.3-Flash, indicating substantial domain-specific optimization. The model outperforms both Claude Fable 5 and Kimi K3, two leading general-purpose AI models that have been widely adopted in legal workflows. At $0.15 per test case, GLM-Legal delivers this advanced performance at a competitive price point that makes it accessible for both large law firms and solo practitioners.

The Harvey Legal Agent Benchmark evaluates models across complex legal tasks including contract analysis, case law research, regulatory compliance assessment, and legal document drafting. GLM-Legal's performance suggests it has been specifically trained on legal corpus data and fine-tuned for the nuanced reasoning patterns required in legal practice. According to Z.ai, the model handles multi-step legal reasoning chains and can navigate the ambiguity inherent in legal interpretation.

Technical Architecture and Base Model

GLM-Legal is built on GLM-5.3-Flash, a lightweight foundation model designed for rapid inference and efficient deployment. The base model provides the computational backbone, while domain-specific training layers add legal expertise. This architecture allows GLM-Legal to maintain fast response times critical for legal research workflows while delivering specialized accuracy on legal tasks.

The model leverages parameter-efficient fine-tuning techniques to adapt GLM-5.3-Flash's general language understanding to legal domain requirements. This approach preserves the base model's efficiency while incorporating the specialized knowledge needed to interpret statutes, precedents, and contractual language. Z.ai has not disclosed the exact training dataset composition, but industry observers note the model's strong performance suggests extensive exposure to case law, legal scholarship, and regulatory documents.

Access Model and Availability

GLM-Legal is currently available through an application-only access model. Legal professionals, law firms, and legal technology companies must submit applications to Z.ai for evaluation before gaining access to the model. This gated approach allows Z.ai to ensure appropriate use cases and gather feedback from qualified users during the initial deployment phase.

The application process requires prospective users to demonstrate their legal credentials and describe their intended use cases. Z.ai reviews applications on a rolling basis and grants access based on alignment with the model's design parameters. This controlled rollout mirrors strategies used by other AI labs launching specialized professional models, balancing broad availability with responsible deployment in high-stakes domains like legal practice.

What This Means

The launch of GLM-Legal represents a significant development in the legal AI landscape, demonstrating that specialized domain models can outperform larger general-purpose systems on vertical-specific benchmarks. For legal professionals, the model offers a new tool for accelerating research, document review, and contract analysis tasks at competitive pricing. The application-based access ensures early adopters are qualified users who can provide meaningful feedback, potentially shaping future iterations of the model. As legal AI adoption accelerates across the profession, GLM-Legal's benchmark performance establishes a new baseline for specialized legal reasoning capabilities.