{"benchmark_id":"kernel-bench-l3","benchmark_name":"Kernel Bench L3","benchmark_description":"Kernel Bench L3 evaluates agentic GPU kernel optimization across 50 problems. Qwen reports two metrics for this benchmark: median per-problem speedup over the PyTorch eager reference and the fraction of problems faster than torch.compile.","max_score":1.0,"categories":["agents","code","systems"],"modality":"text","total_models":1,"entries":[{"rank":1,"model_id":"qwen3.7-max","model_name":"Qwen3.7 Max","organization_name":"Alibaba Cloud / Qwen Team","organization_id":"qwen","benchmark_score":0.96,"normalized_score":0.96,"verified":false,"self_reported":true,"provider_id":"novita","input_cost_per_million":1.25,"output_cost_per_million":3.75,"speed_rps":null,"context_window":1000000,"release_date":"2026-05-19","announcement_date":"2026-05-19","multimodal":false,"param_count":null,"is_new":false}]}