[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"model-minimax-m3":3},[4],{"id":5,"aliases":6,"types":7,"pricing":11,"capabilities":26,"model_capabilities":33,"model_data":106,"introduction":150,"description":151,"title_focus":152,"keyword_extras":153},"minimax-m3",[5],[8,9,10],"openai","anthropic","openai\u002Fresponses\u002Fv1",{"input_tokens_tiers":12},{"tiers":13},[14,20],{"max_tokens":15,"pricing":16},512000,{"input_tokens":17,"output_tokens":18,"cache_read_tokens":19,"cache_write_tokens":17},"2.1","8.4","0.21",{"max_tokens":21,"pricing":22},0,{"input_tokens":23,"output_tokens":24,"cache_read_tokens":25,"cache_write_tokens":23},"4.2","16.8","0.84",{"context_window":27,"max_output":15,"released_at":28,"modalities":29},1048576,"2026-06-01",[30,31,32],"text","image","video",{"groups":34,"highlights":95,"overall":99,"scenario":102,"summary":104,"weakness":105},[35,72,86],{"items":36,"name":71},[37,43,47,52,56,60,64,68],{"detail":38,"method":39,"name":40,"samples":41,"score":42},"","主\u002F客观混合","智能体协同",7,51,{"detail":38,"method":44,"name":45,"samples":46,"score":21},"客观","代码生成",10,{"detail":38,"method":48,"name":49,"samples":50,"score":51},"主观","代码理解与调试",12,82,{"detail":38,"method":48,"name":53,"samples":54,"score":55},"创意写作",17,78,{"detail":38,"method":39,"name":57,"samples":58,"score":59},"信息检索 \u002F RAG",13,53,{"detail":38,"method":48,"name":61,"samples":62,"score":63},"长文本处理",3,90,{"detail":38,"method":44,"name":65,"samples":66,"score":67},"数学与科学推理",14,86,{"detail":38,"method":39,"name":69,"samples":46,"score":70},"多语言能力",67,"专项进阶能力",{"items":73,"name":85},[74,78,81],{"detail":38,"method":39,"name":75,"samples":76,"score":77},"通识知识",15,62,{"detail":38,"method":48,"name":79,"samples":80,"score":63},"文本理解",5,{"detail":38,"method":39,"name":82,"samples":83,"score":84},"逻辑推理",20,79,"通用基础能力",{"items":87,"name":94},[88,91],{"detail":38,"method":48,"name":89,"samples":90,"score":70},"幻觉控制",9,{"detail":38,"method":48,"name":92,"samples":66,"score":93},"指令遵循",75,"安全与合规能力",[96,97,98],{"name":79,"score":63},{"name":61,"score":63},{"name":65,"score":67},{"rating":100,"score":101},"良好",68,{"fit":101,"name":103},"代码研发 \u002F 智能客服 \u002F 内容分析","Minimax M3在本轮能力评测中综合得分68，整体表现良好。其文本理解（90%）和长文本处理（90%）表现突出，能够分析文档语义、提取关键信息并处理复杂问答，并连续理解长篇材料并完成跨段落信息整合。对企业知识库团队、AI 应用开发者和研究人员而言，它适合用于资料检索、文档问答、信息抽取和检索增强生成场景，解决从大量资料中找到相关信息，并形成可用于后续决策的回答；数学与科学推理（86%）也为相关任务提供补充。不过，代码生成（0%）相对较弱，不适合直接承担代码编写任务；如用于研发流程，应限制在代码分析或辅助判断，并由开发者验证结果。",{"name":45,"note":38,"score":21},{"bars":107,"metrics":112,"recentAvailability":138,"summary":139},{"degraded":108,"down":109},[],[21,110,111],1,2,[113,116,119,123,127,131,135],{"label":114,"value":115},"完成率","98%",{"label":117,"value":118},"缓存命中率","28.1%",{"label":120,"unit":121,"value":122},"累计评测耗时","s","8569.6",{"label":124,"unit":125,"value":126},"Token 消耗","tokens","356,338",{"label":128,"unit":129,"value":130},"平均请求次数","次","1.0",{"label":132,"unit":133,"value":134},"TPM","tokens\u002Fmin","2,495",{"label":136,"unit":137,"value":130},"QPM","req\u002Fmin","98.66%",[140,143,146],{"label":141,"unit":121,"value":142},"延迟","19.09",{"label":144,"unit":133,"value":145},"吞吐量","2.5K",{"label":147,"unit":148,"value":149},"可用率","%","98.66","Minimax M3面向文档分析、信息抽取与复杂语义理解提供稳定的模型能力，优势集中在文本理解和长文本处理，并支持图像、视频信息输入。适合需要兼顾任务理解与结果质量的产品和研发团队。","Minimax M3具备文本理解、长文本处理、数学与科学推理能力，可完成文档分析、信息抽取和复杂问答，并支持长篇材料理解和跨段落信息整合。面向企业知识库团队、AI 应用开发者和研究人员，适合资料检索、文档问答、信息抽取和检索增强生成；可通过 SilvaMux 统一 API 一键快速接入。","长文本理解",[79,61]]