[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"model-qwen3.5-flash":3},[4],{"id":5,"aliases":6,"types":7,"pricing":10,"capabilities":49,"model_capabilities":56,"model_data":130,"introduction":173,"description":174,"title_focus":175,"keyword_extras":176},"qwen3.5-flash",[5],[8,9],"openai","anthropic",{"input_tokens_tiers":11,"builtin_tool_calls":34},{"tiers":12},[13,20,27],{"max_tokens":14,"pricing":15},131072,{"input_tokens":16,"output_tokens":17,"cache_read_tokens":18,"cache_write_tokens":19},"0.2","2","0.02","0.25",{"max_tokens":21,"pricing":22},262144,{"input_tokens":23,"output_tokens":24,"cache_read_tokens":25,"cache_write_tokens":26},"0.8","8","0.08","1",{"max_tokens":28,"pricing":29},0,{"input_tokens":30,"output_tokens":31,"cache_read_tokens":32,"cache_write_tokens":33},"1.2","12","0.12","1.5",[35,38,41,43,46],{"tool_name":36,"price_per_unit":37},"web_search","4",{"tool_name":39,"price_per_unit":40},"web_extractor","0",{"tool_name":42,"price_per_unit":40},"code_interpreter",{"tool_name":44,"price_per_unit":45},"t2i_search","24",{"tool_name":47,"price_per_unit":48},"i2i_search","48",{"context_window":50,"released_at":51,"modalities":52},1000000,"2026-02-25",[53,54,55],"text","image","video",{"groups":57,"highlights":119,"overall":123,"scenario":126,"summary":128,"weakness":129},[58,95,109],{"items":59,"name":94},[60,66,71,76,80,84,88,92],{"detail":61,"method":62,"name":63,"samples":64,"score":65},"","主\u002F客观混合","智能体协同",7,79,{"detail":61,"method":67,"name":68,"samples":69,"score":70},"客观","代码生成",10,80,{"detail":61,"method":72,"name":73,"samples":74,"score":75},"主观","代码理解与调试",12,72,{"detail":61,"method":72,"name":77,"samples":78,"score":79},"创意写作",17,61,{"detail":61,"method":62,"name":81,"samples":82,"score":83},"信息检索 \u002F RAG",13,71,{"detail":61,"method":72,"name":85,"samples":86,"score":87},"长文本处理",3,75,{"detail":61,"method":67,"name":89,"samples":90,"score":91},"数学与科学推理",14,93,{"detail":61,"method":62,"name":93,"samples":69,"score":87},"多语言能力","专项进阶能力",{"items":96,"name":108},[97,100,104],{"detail":61,"method":62,"name":98,"samples":99,"score":75},"通识知识",15,{"detail":61,"method":72,"name":101,"samples":102,"score":103},"文本理解",5,85,{"detail":61,"method":62,"name":105,"samples":106,"score":107},"逻辑推理",20,67,"通用基础能力",{"items":110,"name":118},[111,115],{"detail":61,"method":72,"name":112,"samples":113,"score":114},"幻觉控制",9,87,{"detail":61,"method":72,"name":116,"samples":90,"score":117},"指令遵循",77,"安全与合规能力",[120,121,122],{"name":89,"score":91},{"name":112,"score":114},{"name":101,"score":103},{"rating":124,"score":125},"良好",76,{"fit":125,"name":127},"数据分析 \u002F 代码研发辅助 \u002F 智能客服","Qwen3.5 Flash在本轮能力评测中综合得分76，整体表现良好。其数学与科学推理（93%）和幻觉控制（87%）表现突出，能够处理数学求解、科学分析和结构化推理任务，并在知识问答和专业分析中降低无依据内容的风险。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；文本理解（85%）也为相关任务提供补充。使用时应留意，创意写作（61%）相对较弱，在开放式创意和品牌内容任务中，建议由内容人员进一步编辑和筛选。",{"name":77,"note":61,"score":79},{"bars":131,"metrics":134,"recentAvailability":161,"summary":162},{"degraded":132,"down":133},[],[28],[135,138,141,145,149,153,157],{"label":136,"value":137},"完成率","92%",{"label":139,"value":140},"缓存命中率","动态变化",{"label":142,"unit":143,"value":144},"累计评测耗时","s","4745.2",{"label":146,"unit":147,"value":148},"Token 消耗","tokens","4,032,805",{"label":150,"unit":151,"value":152},"平均请求次数","次","12.3",{"label":154,"unit":155,"value":156},"TPM","tokens\u002Fmin","50,992",{"label":158,"unit":159,"value":160},"QPM","req\u002Fmin","3.7","99.7%",[163,166,169],{"label":164,"unit":143,"value":165},"延迟","7.57",{"label":167,"unit":155,"value":168},"吞吐量","50.99K",{"label":170,"unit":171,"value":172},"可用率","%","99.7","Qwen3.5 Flash是一款面向高频处理场景的语言模型，数学推理、幻觉控制和文本理解较为突出，适合批量问答、信息抽取与需要稳定响应的分析任务。","Qwen3.5 Flash具备数学与科学推理、幻觉控制、文本理解能力，可完成数学求解、科学分析和结构化推理，并支持降低知识问答和专业分析中的无依据内容。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。","快速数学模型",[177,178],"数学科学推理","事实可靠性"]