The model costs $0.15 per Harvey's task, roughly 70 times less than the $10.55 required for Gemini 4 Argon. Performance on LegalBench increased by 2.3 percentage points, with a 5 point jump in interpretation tasks, partly due to lower token usage per answer. However, GLM-Legal trailed GLM 5.3 Flash by 4 points on the LegalResearchBench because it searched case law only 6 times per question compared to 16 times for the base model.
Wednesday, October 21, 2026
Story
No. 1 Open Weight Legal AI Debuts in Zai's First Professional Model
AI9 days ago
Zai's new GLM-Legal model scored 19.2% on Harvey's Legal Agent Benchmark, more than double the 8.3% result of the full-size GLM 5.3 and triple the 6.7% of GLM 5.3 Flash. Built on the Flash architecture, the system is now the top open weight model on both LegalBench and Harvey's benchmark, meeting 92.5% of the individual criteria for complex legal assignments involving spreadsheets and presentations. It was designed to handle professional workloads such as evidence analysis, contract review, and compliance.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.