[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"model-kimi-k3":3},[4,23],{"id":5,"aliases":6,"types":7,"pricing":10,"capabilities":15,"priority":22},"kimi-k3",[5],[8,9],"openai","anthropic",{"flat_tokens":11},{"input_tokens":12,"output_tokens":13,"cache_read_tokens":14,"cache_write_tokens":12},"20","100","2",{"context_window":16,"max_output":17,"modalities":18},1048576,131072,[19,20,21],"text","image","video",1,{"id":5,"aliases":24,"types":25,"pricing":26,"capabilities":28,"tier":30,"priority":22,"model_capabilities":31,"model_data":104,"introduction":148,"description":149,"title_focus":150,"keyword_extras":151},[5],[8],{"flat_tokens":27},{"input_tokens":12,"output_tokens":13,"cache_read_tokens":14,"cache_write_tokens":12},{"context_window":16,"max_output":17,"modalities":29},[19,20,21],"tune",{"groups":32,"highlights":94,"overall":98,"scenario":100,"summary":102,"weakness":103},[33,69,84],{"items":34,"name":68},[35,41,46,50,54,58,62,66],{"detail":36,"method":37,"name":38,"samples":39,"score":40},"","主\u002F客观混合","智能体协同",7,75,{"detail":36,"method":42,"name":43,"samples":44,"score":45},"客观","代码生成",10,80,{"detail":36,"method":37,"name":47,"samples":48,"score":49},"代码理解与调试",12,62,{"detail":36,"method":37,"name":51,"samples":52,"score":53},"创意写作",17,60,{"detail":36,"method":37,"name":55,"samples":56,"score":57},"信息检索 \u002F RAG",13,88,{"detail":36,"method":37,"name":59,"samples":60,"score":61},"长文本处理",3,32,{"detail":36,"method":42,"name":63,"samples":64,"score":65},"数学与科学推理",14,86,{"detail":36,"method":37,"name":67,"samples":44,"score":45},"多语言能力","专项进阶能力",{"items":70,"name":83},[71,74,79],{"detail":36,"method":37,"name":72,"samples":73,"score":40},"通识知识",15,{"detail":36,"method":75,"name":76,"samples":77,"score":78},"主观","文本理解",5,96,{"detail":36,"method":37,"name":80,"samples":81,"score":82},"逻辑推理",20,70,"通用基础能力",{"items":85,"name":93},[86,90],{"detail":36,"method":37,"name":87,"samples":88,"score":89},"幻觉控制",9,81,{"detail":36,"method":75,"name":91,"samples":64,"score":92},"指令遵循",85,"安全与合规能力",[95,96,97],{"name":76,"score":78},{"name":55,"score":57},{"name":63,"score":65},{"rating":99,"score":40},"良好",{"fit":40,"name":101},"智能客服 \u002F 内容审核与风控 \u002F 代码研发辅助","Kimi K3在本轮能力评测中综合得分75，整体表现良好。其文本理解（96%）和信息检索\u002FRAG（88%）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，并从知识库和多来源资料中定位信息、组织证据并生成有依据的回答。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。在能力边界上，长文本处理（32%）相对较弱，在超长材料、跨章节引用和持续上下文任务中，建议拆分输入并核对关键信息是否遗漏。",{"name":59,"note":36,"score":61},{"bars":105,"metrics":109,"recentAvailability":136,"summary":137},{"degraded":106,"down":107},[],[108],0,[110,113,116,120,124,128,132],{"label":111,"value":112},"完成率","100%",{"label":114,"value":115},"缓存命中率","动态变化",{"label":117,"unit":118,"value":119},"累计评测耗时","s","8121.2",{"label":121,"unit":122,"value":123},"Token 消耗","tokens","1,019,814",{"label":125,"unit":126,"value":127},"平均请求次数","次","7.0",{"label":129,"unit":130,"value":131},"TPM","tokens\u002Fmin","7,534",{"label":133,"unit":134,"value":135},"QPM","req\u002Fmin","1.2","99.6%",[138,141,144],{"label":139,"unit":118,"value":140},"延迟","22.08",{"label":142,"unit":130,"value":143},"吞吐量","7.53K",{"label":145,"unit":146,"value":147},"可用率","%","99.6","Kimi K3是一款侧重知识检索和推理的语言模型，能够完成资料查找、证据整合与数学分析，适合企业知识库问答、研究辅助及需要检索支撑的复杂任务。","Kimi K3具备文本理解、信息检索\u002FRAG、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，并支持知识库检索、资料问答和证据整合。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。","检索推理模型",[152,153],"信息检索与RAG","数学科学推理"]