Z.ai GLM-Legal Beats Claude Fable 5 and Kimi K3 on Harvey LAB
What happened
Z.ai released GLM-Legal, a legal-focused AI model based on GLM-5.3-Flash, which scored 19.2% all-pass on Harvey’s Legal Agent Benchmark. This score is almost three times higher than its base model’s performance and was achieved at a cost of $0.15 per test. Access to GLM-Legal is by application only.
Why it matters
Legal AI tools still struggle with accuracy and cost-efficiency, so hitting a near 20% all-pass rate shows clear progress in specialized legal models. GLM-Legal’s improvement over its base model signals that targeted fine-tuning can make AI systems more reliable for legal work. The low per-test cost also points to more affordable scaling for firms and developers experimenting with AI in legal contexts. This performance puts pressure on competitors like Claude Fable 5 and Kimi K3, which lag behind on the same benchmark.
What to watch next
Watch for broader adoption of GLM-Legal or similar specialized legal AIs as firms seek more accurate assistants and automated workflows. Its access-by-application approach could indicate controlled release while finalizing safety or compliance checks. Competitors will likely respond with improved models or pricing to retain relevance. Tracking performance changes in Harvey’s Legal Agent Benchmark will show whether this marks a sustained step forward or a temporary advantage.
AI Quick Briefs Editorial Desk