{"benchmark_id":"cfeval","benchmark_name":"CFEval","benchmark_description":"CFEval benchmark for evaluating code generation and problem-solving capabilities","max_score":10000.0,"categories":["code"],"modality":"text","total_models":2,"entries":[{"rank":1,"model_id":"qwen3-235b-a22b-thinking-2507","model_name":"Qwen3-235B-A22B-Thinking-2507","organization_name":"Alibaba Cloud / Qwen Team","organization_id":"qwen","benchmark_score":2134.0,"normalized_score":0.2134,"verified":false,"self_reported":true,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"speed_rps":null,"context_window":null,"release_date":"2025-07-25","announcement_date":"2025-07-25","multimodal":false,"param_count":235000000000,"is_new":false},{"rank":2,"model_id":"qwen3-next-80b-a3b-thinking","model_name":"Qwen3-Next-80B-A3B-Thinking","organization_name":"Alibaba Cloud / Qwen Team","organization_id":"qwen","benchmark_score":2071.0,"normalized_score":0.2071,"verified":false,"self_reported":true,"provider_id":null,"input_cost_per_million":null,"output_cost_per_million":null,"speed_rps":null,"context_window":null,"release_date":"2025-09-10","announcement_date":"2025-09-10","multimodal":false,"param_count":80000000000,"is_new":false}]}