GLM-5.1 Trained Into Top Legal AI Model, Outperforms GPT-5.5
Harvey.ai and Applied Compute fine-tuned GLM-5.1 to surpass GPT-5.5 xhigh and Opus 4.8 Max on legal benchmarks.Summary
Harvey.ai and Applied Compute have transformed Z.ai’s GLM-5.1 into a leading legal AI model by fine-tuning it on proprietary benchmarks. The collaboration achieved a 91.3% rubric pass rate, surpassing GPT-5.5 xhigh and Opus 4.8 Max, while also improving the all-pass rate to 12.6%. The project utilized advanced optimization techniques, including grader alignment, harness optimization, and reinforcement learning, to enhance the model's performance. Despite being a general-purpose model, GLM-5.1's extended context capabilities and tool integration make it highly effective for complex legal workflows. This success suggests that general-purpose models, when fine-tuned with domain-specific data, can rival specialized systems, setting a new standard for the legal AI landscape.
(Source：Blockchain News)
