[{"data":1,"prerenderedAt":198},["ShallowReactive",2],{"copy:zh:site:":3,"copy:zh:models:index,ui":123,"copy:zh:site:common":196},{"common":4,"nav":16,"footer":44,"home":51,"error":116},{"skip":5,"menu":6,"language":7,"sections":8,"footer":9,"tool":10,"home":11,"breadcrumbs":12,"copy":13,"copied":14,"download":15},"跳转到正文","菜单","语言","章节","页脚","工具","首页","面包屑导航","复制","已复制","下载",{"groups":17,"items":25},{"build":18,"connect":19,"instruct":20,"run":21,"web":22,"secure":23,"track":24},"构建","连接","编写指令","运行","开放你的网站","安全","跟踪",{"models":26,"cost":27,"frameworks":28,"mcp":29,"mcpConfig":30,"mcpClients":31,"agentsMd":32,"files":33,"openclaw":34,"agentReady":35,"llmsTxt":36,"robotsTxt":37,"apiCatalog":38,"bots":39,"security":40,"incidents":41,"checklist":42,"changelog":43},"模型与 API 价格","Agent 成本计算器","框架与 SDK","MCP 服务器","MCP 配置生成器","MCP 客户端","AGENTS.md 生成器","指令文件","自托管 OpenClaw","适配 Agent 的网站","llms.txt 生成器","面向 AI 机器人的 robots.txt","API 目录生成器","AI 机器人目录","Agent 安全","事故记录","安全检查清单","更新日志",{"about":45,"privacy":46,"terms":47,"sitemap":48,"disclaimer":49,"checked":50},"关于","隐私","条款","网站地图","为 AI agent 开发者提供的独立参考资料。与此处提及的任何厂商均无关联。","事实核查日期：{date}",{"seo":52,"eyebrow":55,"title":56,"lead":57,"ctaMcp":58,"ctaCost":59,"ctaAgents":60,"sheetLabel":61,"sheetTitle":62,"sheet":63,"sheetNote":69,"sectionsTitle":70,"sections":71,"latestTitle":99,"allChanges":100,"rulesTitle":101,"rules":102,"aboutLink":115},{"title":53,"description":54},"DotsAgent：构建 AI agent 的参考资料与工具","LLM API 价格、agent 成本计算器、适用于 13 个客户端的 MCP 配置、AGENTS.md 和 llms.txt 生成器，以及 agent 安全指南。免费提供，支持 18 种语言。","面向 AI agent 开发者","Agent 开发者的案头参考","涵盖 agent 开发所需的价格、协议、配置文件和安全指南，所有信息均经厂商官方文档核实。查找信息、生成所需文件，然后继续编写代码。","生成 MCP 配置","计算 agent 任务成本","编写 AGENTS.md","最新数据","技术栈概览",{"mcp":64,"openclaw":65,"cheapest":66,"models":67,"servers":29,"bots":68},"MCP 规范","OpenClaw","最低价模型，$\u002F百万输入\u002F输出 tokens","已统计价格的模型","已跟踪的 AI 机器人","每项数据均经人工对照厂商文档核实。","网站内容",{"models":72,"mcp":76,"instructions":79,"frameworks":82,"openclaw":85,"agentReady":88,"security":92,"changelog":95},{"title":73,"body":74,"unit":75},"模型与价格","并列比较每百万输入、输出和缓存 token 的价格，并提供计算器，估算完整 agent 运行的费用。","个模型",{"title":29,"body":77,"unit":78},"Model Context Protocol 服务器目录，以及配置生成器，可为每种客户端生成对应的配置文件。","台服务器",{"title":33,"body":80,"unit":81},"编写 AGENTS.md，并了解各工具如何读取 CLAUDE.md、.cursor\u002Frules 等类似文件。","种文件格式",{"title":28,"body":83,"unit":84},"按语言、许可证、MCP 支持和多 agent 功能对比 agent 框架与厂商 SDK。","个框架",{"title":34,"body":86,"unit":87},"逐步指导你在自己的设备或服务器上安装、配置和更新 OpenClaw。","当前版本",{"title":89,"body":90,"unit":91},"适配 agent 的网站","生成 llms.txt、面向 AI crawler 的 robots.txt 规则和 api-catalog，并通过 user agent 查询任意机器人。","个已收录机器人",{"title":40,"body":93,"unit":94},"了解 agent 面临的 OWASP 风险、真实事件记录，以及上线前可逐项核对的清单。","起已记录事件",{"title":96,"body":97,"unit":98},"API 更新日志","按日期记录模型 API、SDK 和协议的变更，并附上厂商公告链接。","条记录","最新变更","所有变更","我们如何确保信息准确",[103,106,109,112],{"title":104,"body":105},"事实均有来源。","每项价格、版本和 flag 都链接到对应的供应商页面，方便你自行核实。",{"title":107,"body":108},"定期核查。","每个页面都会显示上次核查信息的时间；过时的条目会重新核查或移除。",{"title":110,"body":111},"不接受付费推广。","没有人能付费获得收录、提高排名或得到更好的描述。",{"title":113,"body":114},"工具在浏览器中运行。","生成器和计算器都在本地运行，你输入的内容不会发送到服务器。","关于 DotsAgent",{"title":117,"body":118,"home":119,"popular":120,"failed":121,"failedBody":122},"找不到页面","此地址没有对应的页面。页面可能已迁移，或者链接中有拼写错误。","前往首页","热门工具","出了点问题","我们这边未能加载此页面。请稍后再试，或从首页开始浏览。",{"index":124,"ui":163},{"seo":125,"crumb":26,"eyebrow":128,"title":129,"lead":130,"cheapestLabel":131,"perCodingTask":132,"tableTitle":133,"toCalculator":134,"readTitle":135,"read":136,"faqTitle":149,"faq":150},{"title":126,"description":127},"AI Agent 的 LLM API 定价：28 款模型对比","对比 OpenAI、Anthropic、Google、xAI、DeepSeek、Qwen 等厂商的 28 款模型，列出每百万 token 的输入、缓存和输出价格，以及上下文长度和 Agent 功能。","构建 · 价格核验于 2026 年 10 月 1 日","面向 Agent 开发者的模型与 API 价格","Agent 每执行一步都要付费，而大部分费用都花在之前已发送过的输入上。本表并列展示公开定价、缓存折扣和上下文限制，帮助你根据整个任务的成本选择模型，而不是只看醒目的单价。","单个任务成本最低","完成一个 30 步编码任务的费用，我们的参考工作负载","所有模型，一览无余","估算你自己的工作负载","如何看懂这些价格",[137,140,143,146],{"title":138,"body":139},"缓存输入价格最值得关注","在 Agent 循环中，每次调用都会重新发送 system prompt、工具定义和之前的对话。当前缀从缓存中读取时，服务商只收取常规输入价格的一部分，通常不到十分之一。因此，缓存价格低的模型可能比标价更低的模型更划算。",{"title":141,"body":142},"长 prompt 可能让整次调用涨价","OpenAI、Google 的 Pro 预览版、xAI 和 Qwen 会在单个 prompt 超过 200K 或 272K token 等阈值后提高价格，而且这次请求中的每个 token 都按更高价格计费。目前，Anthropic 对其完整的百万 token 上下文窗口按标准价格收费。",{"title":144,"body":145},"推理 token 按输出计费","回答前会进行思考的模型，其隐藏推理过程按输出价格计费。常规步骤可调低推理强度设置，把高强度设置留给规划或复杂调试。",{"title":147,"body":148},"批处理和非高峰时段折扣通常不适合 Agent","批处理 endpoint 价格减半，但结果可能要几小时才能返回，因此更适合评测和历史数据回填，而非实时 Agent。DeepSeek 的非高峰时段价格是个例外，适用于定时任务。","开发者常见问题",[151,154,157,160],{"q":152,"a":153},"哪个模型最适合低成本运行 AI Agent？","在我们的 30 步编码工作负载中，成本最低的是小型模型：GPT-6 Luna、Qwen3.7-Flash 和 GLM-5.3-Flash，每个任务约需 10 到 20 美分。单步便宜不代表整个任务便宜：如果模型需要更多步骤或重试，成本可能更高。建议用自己的调用记录测试两三个候选模型。",{"q":155,"a":156},"为什么我的 Agent 费用远高于按 token 单价估算的金额？","因为每一步都会重新发送不断增长的对话内容。一个起始 prompt 有 25K token 的 30 步任务，总输入量会远超 300 万 token。与更换模型相比，使用 prompt 缓存、缩短工具输出和总结较早的对话轮次更能降低成本。",{"q":158,"a":159},"这些是最终价格吗？","这些是各服务商定价页面在 2026 年 10 月 1 日公布的美元标价，不含税费、地区附加费或协商折扣。标注为第三方数据的价格无法从厂商自己的页面获取。",{"q":161,"a":162},"缓存读取和缓存写入有什么区别？","缓存读取是读取缓存中已有 token 时享受的折扣价格。一些服务商（包括 Anthropic）还会在前缀首次写入缓存时收取额外费用。表格列出的是读取价格；适用的写入价格会在各模型页面中注明。",{"crumb":164,"sources":165,"filters":166,"tiers":171,"table":175,"flags":184,"caps":188},"模型","来源",{"provider":167,"tier":168,"all":169,"anyTier":170},"按提供商筛选","按级别筛选","所有提供商","所有级别",{"flagship":172,"mid":173,"small":174},"旗舰","主力","小型高速",{"caption":176,"model":164,"provider":177,"input":178,"cached":179,"output":180,"context":181,"released":182,"perM":183},"用于构建智能体的模型 API 标价","提供商","输入","缓存输入","输出","上下文","发布时间","每百万 token 的美元价格；标准级别、短 prompt 费率",{"preview":185,"thirdParty":186,"long":187},"预览版","第三方价格","长 prompt 费用更高",{"tools":189,"structured":190,"computerUse":191,"mcp":192,"webSearch":193,"reasoning":194,"vision":195},"工具调用","结构化输出","计算机操作","远程 MCP","内置网页搜索","可调节推理","图像输入",{"common":197},{"skip":5,"menu":6,"language":7,"sections":8,"footer":9,"tool":10,"home":11,"breadcrumbs":12,"copy":13,"copied":14,"download":15},1790863379068]