[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"api:\u002Fnav:{}":3,"api:\u002Ftopics:{}":104},{"login_code_required":4,"data_date":5,"vendors":6,"categories":39,"topics":55,"model_count":72,"benchmark_count":73,"models":74,"benchmarks":79},false,"2026-10-27",[7,15,23,31],{"name":8,"slug":9,"models":10},"阿里巴巴","alibaba",[11],{"name":12,"slug":13,"short_name":12,"overall_score":14},"Qwen3.8-27B","qwen3-8-27b","47.576875000000000000000000",{"name":16,"slug":17,"models":18},"深度求索","deepseek",[19],{"name":20,"slug":21,"short_name":20,"overall_score":22},"DeepSeek-V4-Pro","deepseek-v4-pro","65.030000000000000000000000",{"name":24,"slug":25,"models":26},"月之暗面","moonshot",[27],{"name":28,"slug":29,"short_name":28,"overall_score":30},"Kimi-K3","kimi-k3","57.474000000000000000000000",{"name":32,"slug":33,"models":34},"智谱","zhipu",[35],{"name":36,"slug":37,"short_name":36,"overall_score":38},"GLM-5.3","glm-5-3","61.067500000000000000000000",[40,44,48,51],{"name":41,"slug":42,"benchmark_count":43},"体验感","experience",2,{"name":45,"slug":46,"benchmark_count":47},"认知能力","cognition",5,{"name":49,"slug":50,"benchmark_count":43},"Agentic 能力","agentic",{"name":52,"slug":53,"benchmark_count":54},"应用领域","domain",1,[56,60,64,68],{"slug":57,"name":58,"question":59},"quant","量化 vs 原精度","量化之后能力掉多少",{"slug":61,"name":62,"question":63},"selfhost","自部署 vs 官方 API","同一份权重，自己部署和调官方 API，分数一致吗",{"slug":65,"name":66,"question":67},"size","同家族尺寸","同一代模型，小尺寸比旗舰差多少",{"slug":69,"name":70,"question":71},"harness","多 Harness","同一个模型换一个 harness，agentic 分数变多少",4,8,[75,76,77,78],{"name":20,"slug":21},{"name":36,"slug":37},{"name":28,"slug":29},{"name":12,"slug":13},[80,83,86,89,92,95,98,101],{"name":81,"slug":82},"AA-LCR","aa-lcr",{"name":84,"slug":85},"AA-Omniscience","aa-omniscience",{"name":87,"slug":88},"AutomationBench","automationbench",{"name":90,"slug":91},"CL-bench","cl-bench",{"name":93,"slug":94},"HLE","hle",{"name":96,"slug":97},"MMLU-Pro","mmlu-pro",{"name":99,"slug":100},"SimpleQA Verified","simpleqa-verified",{"name":102,"slug":103},"SWE-bench Pro","swe-bench-pro",{"topics":105},[106,130,184,210],{"slug":57,"name":58,"question":59,"method":107,"kind":108,"groups":109,"comparison_count":125,"average_delta":126,"max_absolute_delta":126,"largest_gap_groups":127,"within_one_point":125,"category_average_deltas":128,"largest_drop_category":126,"breakdown":129,"selected_summary":126},"同一份权重、同一推理引擎和硬件，只换精度。基线是官方 BF16，全部自部署，最高推理档。","variants",[110,113,117,121],{"name":12,"baseline":111,"rows":112,"complete":4},"qwen3-8-27b-bf16-max",[],{"name":114,"baseline":115,"rows":116,"complete":4},"Qwen3.8-Flash-Next","qwen3-8-flash-next-bf16-max",[],{"name":118,"baseline":119,"rows":120,"complete":4},"GLM-5.3-Flash","glm-5-3-flash-bf16-max",[],{"name":122,"baseline":123,"rows":124,"complete":4},"MiniMax-M3","minimax-m3-bf16-max",[],0,null,[],{},{},{"slug":61,"name":62,"question":63,"method":131,"kind":108,"groups":132,"comparison_count":125,"average_delta":126,"max_absolute_delta":126,"largest_gap_groups":181,"within_one_point":125,"category_average_deltas":182,"largest_drop_category":126,"breakdown":183,"selected_summary":126},"自部署用官方发布的权重和推荐配置，官方 API 用厂商默认精度。两边都是最高档，评测配置相同。",[133,157,161,178],{"name":134,"baseline":135,"rows":136,"complete":4},"Kimi K3","kimi-k3-api-max",[137],{"access":138,"effort":139,"family":28,"is_tier":140,"hardware":126,"model_id":141,"provider":126,"data_date":142,"precision":126,"categories":143,"is_primary":140,"model_name":28,"model_slug":29,"short_name":28,"variant_id":151,"effort_rank":152,"input_price":126,"vendor_name":24,"vendor_slug":25,"architecture":126,"output_price":126,"overall_rank":141,"ranked_count":72,"release_date":153,"variant_name":154,"variant_slug":135,"vendor_order":125,"weights_repo":126,"missing_count":141,"overall_score":30,"context_window":126,"total_cost_usd":126,"total_params_b":126,"weights_source":126,"active_params_b":126,"cache_hit_price":126,"cost_incomplete":140,"is_open_weights":140,"inference_engine":126,"delta":155,"category_deltas":156,"baseline":140,"cost_ratio":126,"group":134},"api","max",true,3,"2026-10-24",[144,146,148,150],{"name":41,"rank":43,"slug":42,"score":145,"expected":43,"available":54,"ranked_count":72},"65.051903114186850000000000",{"name":45,"rank":43,"slug":46,"score":147,"expected":47,"available":72,"ranked_count":72},"67.050833333333332500000000",{"name":49,"rank":72,"slug":50,"score":149,"expected":43,"available":54,"ranked_count":72},"3.500000000000000000000000",{"name":52,"rank":72,"slug":53,"score":149,"expected":54,"available":54,"ranked_count":72},11,6,"2026-07-16","Kimi-K3 官方 API max","0E-24",{"experience":155,"cognition":155,"agentic":155,"domain":155},{"name":158,"baseline":159,"rows":160,"complete":4},"DeepSeek V4 Pro","deepseek-v4-pro-api-max",[],{"name":36,"baseline":162,"rows":163,"complete":4},"glm-5-3-api-max",[164],{"access":138,"effort":139,"family":36,"is_tier":140,"hardware":126,"model_id":54,"provider":126,"data_date":165,"precision":126,"categories":166,"is_primary":140,"model_name":36,"model_slug":37,"short_name":36,"variant_id":73,"effort_rank":152,"input_price":126,"vendor_name":32,"vendor_slug":33,"architecture":126,"output_price":126,"overall_rank":43,"ranked_count":72,"release_date":175,"variant_name":176,"variant_slug":162,"vendor_order":125,"weights_repo":126,"missing_count":43,"overall_score":38,"context_window":126,"total_cost_usd":126,"total_params_b":126,"weights_source":126,"active_params_b":126,"cache_hit_price":126,"cost_incomplete":140,"is_open_weights":140,"inference_engine":126,"delta":155,"category_deltas":177,"baseline":140,"cost_ratio":126,"group":36},"2026-10-23",[167,169,171,173],{"name":41,"rank":54,"slug":42,"score":168,"expected":43,"available":54,"ranked_count":72},"72.997032640949555000000000",{"name":45,"rank":141,"slug":46,"score":170,"expected":47,"available":72,"ranked_count":72},"59.363333333333333750000000",{"name":49,"rank":54,"slug":50,"score":172,"expected":43,"available":43,"ranked_count":72},"54.225000000000000000000000",{"name":52,"rank":54,"slug":53,"score":174,"expected":54,"available":54,"ranked_count":72},"52.500000000000000000000000","2026-08-18","GLM-5.3 官方 API max",{"experience":155,"cognition":155,"agentic":155,"domain":155},{"name":122,"baseline":179,"rows":180,"complete":4},"minimax-m3-api-max",[],[],{},{},{"slug":65,"name":66,"question":67,"method":185,"kind":108,"groups":186,"comparison_count":125,"average_delta":126,"max_absolute_delta":126,"largest_gap_groups":207,"within_one_point":125,"category_average_deltas":208,"largest_drop_category":126,"breakdown":209,"selected_summary":126},"同一家族、同一代的模型，全部用官方 API 或官方权重，最高档，评测配置相同。",[187,191,200,204],{"name":188,"baseline":189,"rows":190,"complete":4},"Qwen3.8","qwen3-8-max-api-max",[],{"name":36,"baseline":162,"rows":192,"complete":4},[193],{"access":138,"effort":139,"family":36,"is_tier":140,"hardware":126,"model_id":54,"provider":126,"data_date":165,"precision":126,"categories":194,"is_primary":140,"model_name":36,"model_slug":37,"short_name":36,"variant_id":73,"effort_rank":152,"input_price":126,"vendor_name":32,"vendor_slug":33,"architecture":126,"output_price":126,"overall_rank":43,"ranked_count":72,"release_date":175,"variant_name":176,"variant_slug":162,"vendor_order":125,"weights_repo":126,"missing_count":43,"overall_score":38,"context_window":126,"total_cost_usd":126,"total_params_b":126,"weights_source":126,"active_params_b":126,"cache_hit_price":126,"cost_incomplete":140,"is_open_weights":140,"inference_engine":126,"delta":155,"category_deltas":199,"baseline":140,"cost_ratio":126,"group":36},[195,196,197,198],{"name":41,"rank":54,"slug":42,"score":168,"expected":43,"available":54,"ranked_count":72},{"name":45,"rank":141,"slug":46,"score":170,"expected":47,"available":72,"ranked_count":72},{"name":49,"rank":54,"slug":50,"score":172,"expected":43,"available":43,"ranked_count":72},{"name":52,"rank":54,"slug":53,"score":174,"expected":54,"available":54,"ranked_count":72},{"experience":155,"cognition":155,"agentic":155,"domain":155},{"name":201,"baseline":202,"rows":203,"complete":4},"MiMo-V2.6","mimo-v2-6-pro-api-max",[],{"name":205,"baseline":159,"rows":206,"complete":4},"DeepSeek V4",[],[],{},{},{"slug":69,"name":70,"question":71,"method":211,"kind":69,"groups":212,"comparison_count":125,"average_delta":126,"max_absolute_delta":126,"largest_gap_groups":223,"within_one_point":125,"category_average_deltas":224,"largest_drop_category":126,"breakdown":225,"selected_summary":126},"在 SWE-bench Pro 上，同一模型、同一推理档，只换 harness，其余评测配置相同。",[213,216,219,221],{"name":134,"baseline":214,"rows":215,"complete":4},"openhands",[],{"name":217,"baseline":214,"rows":218,"complete":4},"Qwen3.8 Max",[],{"name":36,"baseline":214,"rows":220,"complete":4},[],{"name":158,"baseline":214,"rows":222,"complete":4},[],[],{},{}]