[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"model-deepseek-v4-flash":3},[4,76],{"id":5,"aliases":6,"types":7,"pricing":11,"capabilities":16,"priority":22,"model_capabilities":23,"model_data":25,"introduction":70,"description":71,"title_focus":72,"keyword_extras":73},"deepseek-v4-flash",[5],[8,9,10],"openai","anthropic","openai\u002Fresponses\u002Fv1",{"flat_tokens":12},{"input_tokens":13,"output_tokens":14,"cache_read_tokens":15,"cache_write_tokens":13},"3","9","0.1",{"context_window":17,"max_output":18,"released_at":19,"modalities":20},1048576,384000,"2026-04-24",[21],"text",1,{"summary":24},"DeepSeek V4 Flash在本轮能力评测中综合得分79，整体表现良好。其信息检索 \u002F RAG（94%）和文本理解（92%）表现突出，能够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，并分析文档语义、提取关键信息并处理复杂问答。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；多语言能力（89%）也为相关任务提供补充。使用时应留意，智能体协同（36%）相对较弱，在需要多步骤自主规划、连续工具调用或跨智能体协作的任务中，建议增加人工复核或专项验证。",{"bars":26,"metrics":31,"recentAvailability":58,"summary":59},{"degraded":27,"down":29},[28],2,[30,22],0,[32,35,38,42,46,50,54],{"label":33,"value":34},"完成率","78%",{"label":36,"value":37},"缓存命中率","动态变化",{"label":39,"unit":40,"value":41},"累计评测耗时","s","17803.2",{"label":43,"unit":44,"value":45},"Token 消耗","tokens","1,743,969",{"label":47,"unit":48,"value":49},"平均请求次数","次","5.6",{"label":51,"unit":52,"value":53},"TPM","tokens\u002Fmin","5,877",{"label":55,"unit":56,"value":57},"QPM","req\u002Fmin","0.7","98.8%",[60,63,66],{"label":61,"unit":40,"value":62},"延迟","8.32",{"label":64,"unit":52,"value":65},"吞吐量","5.88K",{"label":67,"unit":68,"value":69},"可用率","%","98.8","DeepSeek V4 Flash是一款侧重信息检索\u002FRAG的模型，在文本理解方面也有较好表现。它适合处理企业知识检索、资料问答与检索增强生成，可作为相关业务流程中的理解、推理或生成能力底座。","DeepSeek V4 Flash具备信息检索\u002FRAG、文本理解、多语言能力，可完成知识库检索、资料问答和证据整合，并支持文档分析、信息抽取和复杂问答。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。","检索增强模型",[74,75],"信息检索与RAG","文本理解",{"id":5,"aliases":77,"types":78,"pricing":79,"capabilities":81,"tier":84,"priority":22,"model_capabilities":85,"model_data":157,"introduction":70,"description":71,"title_focus":72,"keyword_extras":183},[5],[8,9],{"flat_tokens":80},{"input_tokens":13,"output_tokens":14,"cache_read_tokens":15,"cache_write_tokens":13},{"context_window":17,"max_output":82,"released_at":19,"modalities":83},65536,[21],"tune",{"groups":86,"highlights":147,"overall":151,"scenario":154,"summary":24,"weakness":156},[87,124,138],{"items":88,"name":123},[89,95,100,104,108,112,116,120],{"detail":90,"method":91,"name":92,"samples":93,"score":94},"","主\u002F客观混合","智能体协同",7,36,{"detail":90,"method":96,"name":97,"samples":98,"score":99},"客观","代码生成",10,80,{"detail":90,"method":101,"name":102,"samples":103,"score":99},"主观","代码理解与调试",12,{"detail":90,"method":101,"name":105,"samples":106,"score":107},"创意写作",17,83,{"detail":90,"method":91,"name":109,"samples":110,"score":111},"信息检索 \u002F RAG",13,94,{"detail":90,"method":101,"name":113,"samples":114,"score":115},"长文本处理",3,88,{"detail":90,"method":96,"name":117,"samples":118,"score":119},"数学与科学推理",14,86,{"detail":90,"method":91,"name":121,"samples":98,"score":122},"多语言能力",89,"专项进阶能力",{"items":125,"name":137},[126,130,133],{"detail":90,"method":91,"name":127,"samples":128,"score":129},"通识知识",15,81,{"detail":90,"method":101,"name":75,"samples":131,"score":132},5,92,{"detail":90,"method":91,"name":134,"samples":135,"score":136},"逻辑推理",20,70,"通用基础能力",{"items":139,"name":146},[140,144],{"detail":90,"method":101,"name":141,"samples":142,"score":143},"幻觉控制",9,72,{"detail":90,"method":101,"name":145,"samples":118,"score":129},"指令遵循","安全与合规能力",[148,149,150],{"name":109,"score":111},{"name":75,"score":132},{"name":121,"score":122},{"rating":152,"score":153},"良好",79,{"fit":153,"name":155},"代码研发 \u002F 内容创作",{"name":92,"note":90,"score":94},{"bars":158,"metrics":161,"recentAvailability":175,"summary":176},{"degraded":159,"down":160},[],[30,22,28],[162,164,166,168,170,172,174],{"label":33,"value":163},"97%",{"label":36,"value":165},"36.2%",{"label":39,"unit":40,"value":167},"8793.7",{"label":43,"unit":44,"value":169},"241,733",{"label":47,"unit":48,"value":171},"1.0",{"label":51,"unit":52,"value":173},"1,649",{"label":55,"unit":56,"value":171},"98.66%",[177,179,181],{"label":61,"unit":40,"value":178},"12.48",{"label":64,"unit":52,"value":180},"1.65K",{"label":67,"unit":68,"value":182},"98.66",[74,75]]