[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"model-image-service-v2-gpt-5.3-codex":3,"text-service-v2-gpt-5.3-codex":194},[4],{"id":5,"aliases":6,"types":7,"pricing_group":9,"pricing":10,"display_name":5,"model_capabilities":15,"model_capabilities_zh":90,"model_data":139,"introduction":182,"introduction_zh":183,"description":184,"description_zh":185,"title_focus":186,"title_focus_zh":187,"keyword_extras":188,"keyword_extras_zh":191},"gpt-5.3-codex",[5],[8],"openai\u002Fresponses\u002Fv1","openai\u002Fgpt",{"flat_tokens":11},{"input_tokens":12,"output_tokens":13,"cache_read_tokens":14,"cache_write_tokens":12},"1.75","14","0.175",{"groups":16,"highlights":80,"overall":84,"scenario":86,"summary":88,"weakness":89},[17,55,70],{"items":18,"name":54},[19,25,30,35,39,43,47,51],{"detail":20,"method":21,"name":22,"samples":23,"score":24},"","Mixed","Agent Collaboration",7,43,{"detail":20,"method":26,"name":27,"samples":28,"score":29},"Objective","Code Generation",10,99,{"detail":20,"method":31,"name":32,"samples":33,"score":34},"Subjective","Code Comprehension & Debugging",12,82,{"detail":20,"method":31,"name":36,"samples":37,"score":38},"Creative Writing",17,80,{"detail":20,"method":21,"name":40,"samples":41,"score":42},"Information Retrieval \u002F RAG",13,92,{"detail":20,"method":31,"name":44,"samples":45,"score":46},"Long-Context Handling",3,72,{"detail":20,"method":26,"name":48,"samples":49,"score":50},"Math & Science Reasoning",14,36,{"detail":20,"method":21,"name":52,"samples":28,"score":53},"Multilingual Capability",76,"Advanced Capabilities",{"items":56,"name":69},[57,61,65],{"detail":20,"method":21,"name":58,"samples":59,"score":60},"General Knowledge",15,74,{"detail":20,"method":31,"name":62,"samples":63,"score":64},"Text Comprehension",5,86,{"detail":20,"method":21,"name":66,"samples":67,"score":68},"Logical Reasoning",20,46,"General Foundations",{"items":71,"name":79},[72,76],{"detail":20,"method":31,"name":73,"samples":74,"score":75},"Hallucination Control",9,71,{"detail":20,"method":31,"name":77,"samples":49,"score":78},"Instruction Following",78,"Safety & Compliance",[81,82,83],{"name":27,"score":29},{"name":40,"score":42},{"name":62,"score":64},{"rating":85,"score":46},"良好",{"fit":46,"name":87},"Software Engineering \u002F Information Retrieval \u002F Enterprise Knowledge Base","GPT-5.3 Codex scored 72 overall, performing well across the board — a coding-agent model built for real repositories. It stands out in code generation (99%), information retrieval \u002F RAG (92%) and text comprehension (86%), reliably producing working code from a codebase and grounding answers in retrieved documents, which makes it a strong fit for software engineering and enterprise knowledge-base workflows. Comparatively, math and science reasoning (36%) falls short, so quantitative-heavy tasks may need extra verification. The deeper the codebase work, the more value it delivers.",{"name":48,"note":20,"score":50},{"groups":91,"highlights":130,"overall":134,"scenario":135,"summary":137,"weakness":138},[92,114,123],{"items":93,"name":113},[94,97,100,103,105,107,109,111],{"detail":20,"method":95,"name":96,"samples":23,"score":24},"主\u002F客观混合","智能体协同",{"detail":20,"method":98,"name":99,"samples":28,"score":29},"客观","代码生成",{"detail":20,"method":101,"name":102,"samples":33,"score":34},"主观","代码理解与调试",{"detail":20,"method":101,"name":104,"samples":37,"score":38},"创意写作",{"detail":20,"method":95,"name":106,"samples":41,"score":42},"信息检索 \u002F RAG",{"detail":20,"method":101,"name":108,"samples":45,"score":46},"长文本处理",{"detail":20,"method":98,"name":110,"samples":49,"score":50},"数学与科学推理",{"detail":20,"method":95,"name":112,"samples":28,"score":53},"多语言能力","专项进阶能力",{"items":115,"name":122},[116,118,120],{"detail":20,"method":95,"name":117,"samples":59,"score":60},"通识知识",{"detail":20,"method":101,"name":119,"samples":63,"score":64},"文本理解",{"detail":20,"method":95,"name":121,"samples":67,"score":68},"逻辑推理","通用基础能力",{"items":124,"name":129},[125,127],{"detail":20,"method":101,"name":126,"samples":74,"score":75},"幻觉控制",{"detail":20,"method":101,"name":128,"samples":49,"score":78},"指令遵循","安全与合规能力",[131,132,133],{"name":99,"score":29},{"name":106,"score":42},{"name":119,"score":64},{"rating":85,"score":46},{"fit":46,"name":136},"代码研发 \u002F 信息检索 \u002F 企业知识库","GPT-5.3 Codex 综合得分 72，整体表现良好，是一款面向真实代码库的编程智能体模型。代码生成（99%）、信息检索 \u002F RAG（92%）与文本理解（86%）表现突出，能围绕代码库产出可用实现并基于检索资料给出有据回答，在代码研发与企业知识库场景中优势明显。相对而言，数学与科学推理（36%）明显不足，数值密集型任务需谨慎使用。代码库越深的工作越能体现它的价值。",{"name":110,"note":20,"score":50},{"bars":140,"metrics":143,"recentAvailability":170,"summary":171},{"degraded":141,"down":142},[],[],[144,147,150,154,158,162,166],{"label":145,"value":146},"Completion Rate","100%",{"label":148,"value":149},"Cache Hit Rate","0.0%",{"label":151,"unit":152,"value":153},"Total Eval Time","s","1274.9",{"label":155,"unit":156,"value":157},"Token Usage","tokens","85,254",{"label":159,"unit":160,"value":161},"Avg. Requests","次","1.0",{"label":163,"unit":164,"value":165},"TPM","tokens\u002Fmin","4,012",{"label":167,"unit":168,"value":169},"QPM","req\u002Fmin","7.0","100.0%",[172,175,178],{"label":173,"unit":152,"value":174},"Latency","6.53",{"label":176,"unit":164,"value":177},"Throughput","4.01K",{"label":179,"unit":180,"value":181},"Availability","%","100.0","GPT-5.3 Codex is a coding-agent model that analyzes, implements, tests and fixes issues around real codebases — built for software teams that want a model to continuously operate in a development environment.","GPT-5.3 Codex是一款编程智能体模型，能够围绕真实代码库完成分析、实现、测试和问题修复，适合需要让模型持续操作开发环境的软件团队。","GPT-5.3 Codex focuses on software engineering and agentic coding: it understands codebases, edits files and runs development toolchains. Integrate in one click via the SilvaMux unified API.","GPT-5.3 Codex专注软件工程与智能体编程，可理解代码库、修改文件并执行开发工具链。可通过 SilvaMux 统一 API 一键快速接入。","Coding Agent Model","编程智能体模型",[189,190],"codebase understanding","software engineering",[192,193],"代码库理解","软件工程",{"ttft":195,"tps":198,"e2e":200,"requests":203,"successfulRequests":204,"successRate":205,"availability":206},{"p50":196,"p95":197},2032,2398,{"p50":34,"p95":199},97,{"p50":201,"p95":202},3528,7964,191538,189348,"98.86",[207,208,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207,207],"up","degraded"]