[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-claude-api-model-guide-new-top-tier-zh":3,"article-related-claude-api-model-guide-new-top-tier-zh":35,"series-model-release-eee9cee3-8c3d-48a7-b647-98cf19955e54":88},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":27,"views":31,"created_at":32,"published_at":33,"topic_cluster_id":34},"eee9cee3-8c3d-48a7-b647-98cf19955e54","claude-api-model-guide-new-top-tier-zh","Claude API 模型指南升級","\u003Cp data-speakable=\"summary\">\u003Ca href=\"\u002Fnews\u002Fanthropic-pwc-deal-enterprise-ai-adoption-zh\">Anth\u003C\u002Fa>ropic 的 \u003Ca href=\"\u002Ftag\u002Fclaude\">Claude\u003C\u002Fa> 文件把 \u003Ca href=\"\u002Ftag\u002Fopus-47\">Opus 4.7\u003C\u002Fa> 放到最上層，並把模型 ID、價格、上下文與雲端端點一次講清楚。\u003C\u002Fp>\u003Cp>說真的，這頁很像文件版的選購指南。\u003Ca href=\"https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fabout-claude\u002Fmodels\u002Foverview\" target=\"_blank\" rel=\"noopener\">Claude models overview\u003C\u002Fa> 直接把 \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.7\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude Sonnet 4.6\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude Haiku 4.5\u003C\u002Fa> 排好。你不用再翻半天文件猜哪個適合。\u003C\u002Fp>\u003Cp>更實際的是，\u003Ca href=\"\u002Ftag\u002Fanthropic\">Anthropic\u003C\u002Fa> 把 \u003Ca href=\"https:\u002F\u002Faws.amazon.com\u002Fbedrock\u002Fclaude\u002F\" target=\"_blank\" rel=\"noopener\">Amazon Bedrock\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fcloud.google.com\u002Fvertex-ai\" target=\"_blank\" rel=\"noopener\">Vertex AI\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai-foundry\" target=\"_blank\" rel=\"noopener\">Microsoft Foundry\u003C\u002Fa> 的差異也寫進去。對開發者來說，這種資訊比廣告詞有用多了。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Model\u003C\u002Fth>\u003Cth>API ID\u003C\u002Fth>\u003Cth>Input price\u003C\u002Fth>\u003Cth>Output price\u003C\u002Fth>\u003Cth>Context window\u003C\u002Fth>\u003Cth>Max output\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>Claude Opus 4.7\u003C\u002Ftd>\u003Ctd>claude-opus-4-7\u003C\u002Ftd>\u003Ctd>$5 \u002F MTok\u003C\u002Ftd>\u003Ctd>$25 \u002F MTok\u003C\u002Ftd>\u003Ctd>1M tokens\u003C\u002Ftd>\u003Ctd>128k tokens\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Claude Sonnet 4.6\u003C\u002Ftd>\u003Ctd>claude-sonnet-4-6\u003C\u002Ftd>\u003Ctd>$3 \u002F MTok\u003C\u002Ftd>\u003Ctd>$15 \u002F MTok\u003C\u002Ftd>\u003Ctd>1M tokens\u003C\u002Ftd>\u003Ctd>64k tokens\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Claude Haiku 4.5\u003C\u002Ftd>\u003Ctd>claude-haiku-4-5-20251001\u003C\u002Ftd>\u003Ctd>$1 \u002F MTok\u003C\u002Ftd>\u003Ctd>$5 \u002F MTok\u003C\u002Ftd>\u003Ctd>200k tokens\u003C\u002Ftd>\u003Ctd>64k tokens\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Opus 4.7 被放在最前面，不是偶然\u003C\u002Fh2>\u003Cp>\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude Opus 4.7\u003C\u002Fa> 是 Anthropic 目前主推的高階模型。文件直接把它寫成最適合複雜推理和 \u003Ca href=\"\u002Ftag\u002Fagentic-coding\">agentic coding\u003C\u002Fa> 的選項。這種寫法很直白，也很少見。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779023645425-mkdf.png\" alt=\"Claude API 模型指南升級\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Anthropic 還提到，Opus 4.7 相比 Opus 4.6 在 agentic coding 有「step-change improvement」。講白了，就是它想讓你知道，這次升級不是小修小補。對寫程式、跑工具鏈、做長流程任務的人，這句話很有份量。\u003C\u002Fp>\u003Cp>更重要的是，Anthropic 已經把選模邏輯講得很清楚。你要長上下文，就看 1M \u003Ca href=\"\u002Ftag\u002Ftoken\">token\u003C\u002Fa>。你要更強推理，就看 Opus。你要速度和成本平衡，就看 Son\u003Ca href=\"\u002Fnews\u002Fwhy-hpe-greenlake-kubernetes-push-right-move-zh\">net\u003C\u002Fa>。你要低延遲，就看 Haiku。\u003C\u002Fp>\u003Cul>\u003Cli>Opus 4.7：1M token 上下文，128k 輸出，$5 \u002F MTok 輸入，$25 \u002F MTok 輸出\u003C\u002Fli>\u003Cli>Sonnet 4.6：1M token 上下文，64k 輸出，$3 \u002F MTok 輸入，$15 \u002F MTok 輸出\u003C\u002Fli>\u003Cli>Haiku 4.5：200k token 上下文，64k 輸出，$1 \u002F MTok 輸入，$5 \u002F MTok 輸出\u003C\u002Fli>\u003Cli>Opus 和 Sonnet 都支援 1M token，長文件差距沒以前那麼大\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>模型 ID 變得更硬，對 production 是好事\u003C\u002Fh2>\u003Cp>Anthropic 這次把 m\u003Ca href=\"\u002Fnews\u002Fwhy-openai-is-right-to-put-codex-on-phones-zh\">ode\u003C\u002Fa>l ID 的規則講得更細。文件說，每個 Claude model ID 都是 pinned snapshot。只要 ID 裡有日期，它就是固定版本。從 4.6 世代開始，連沒有日期的 ID 也不是永遠指向最新，而是固定快照。\u003C\u002Fp>\u003Cp>這對 production 很重要。你不會希望今天測過的模型，明天偷偷換行為。特別是客服機器人、法務摘要、程式碼生成這種場景，版本漂移常常比模型貴。\u003C\u002Fp>\u003Cp>Anthropic 也把跨平台 ID 規則講清楚。\u003Ca href=\"https:\u002F\u002Faws.amazon.com\u002Fbedrock\u002F\" target=\"_blank\" rel=\"noopener\">AWS Bedrock\u003C\u002Fa> 會用自己的命名法。\u003Ca href=\"https:\u002F\u002Fcloud.google.com\u002Fvertex-ai\" target=\"_blank\" rel=\"noopener\">Vertex AI\u003C\u002Fa> 也有自己的 endpoint 規則。只有 Claude Platform on AWS 會盡量維持和第一方 API 一樣的 model ID。\u003C\u002Fp>\u003Cblockquote>“The best way to use a foundation model is to pick the smallest one that can do the job.” — Dario Amodei, Anthropic CEO\u003C\u002Fblockquote>\u003Cp>這句話放在這篇文件旁邊很合理。Anthropic 沒有叫你全都上最大模型。它是在逼你做選擇。對工程團隊來說，這反而比較健康，因為每個功能都能對應不同模型，而不是全部綁死在同一顆。\u003C\u002Fp>\u003Ch2>雲端端點才是很多團隊真正卡關的地方\u003C\u002Fh2>\u003Cp>很多人只看模型名稱，卻忽略部署地點。Anthropic 這次把 Claude 放在多個雲端平台上，包含 \u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude API\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Faws.amazon.com\u002Fbedrock\u002Fclaude\u002F\" target=\"_blank\" rel=\"noopener\">Amazon Bedrock\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fcloud.google.com\u002Fvertex-ai\" target=\"_blank\" rel=\"noopener\">Vertex AI\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.microsoft.com\u002Fen-us\u002Fai-foundry\" target=\"_blank\" rel=\"noopener\">Microsoft Foundry\u003C\u002Fa>。這不是純技術問題，還牽涉採購、法規、區域限制。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779023638987-pil4.png\" alt=\"Claude API 模型指南升級\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>文件也補了 endpoint 的差異。從 Sonnet 4.5 開始，Bedrock 有 global endpoints 和 regional endpoints。Vertex AI 還多了 multi-region endpoints。這些選項聽起來很枯燥，但對金融、醫療、政府案子來說，這就是能不能上線的差別。\u003C\u002Fp>\u003Cp>還有一個很實用的細節。文件說 Max output 是針對同步 Messages API。某些模型在 Message Batches API 下，搭配 beta header，還能拉更高。這種資訊很像藏在角落的備註，但真的會影響架構設計。\u003C\u002Fp>\u003Cul>\u003Cli>Claude Platform on AWS 直接沿用第一方 Claude API IDs\u003C\u002Fli>\u003Cli>Bedrock 提供 global 與 regional endpoints\u003C\u002Fli>\u003Cli>Vertex AI 提供 global、multi-region、regional endpoints\u003C\u002Fli>\u003Cli>Models API 會回傳 max_input_tokens、max_tokens、capabilities\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>和其他模型比，Anthropic 的路線更像工程文件\u003C\u002Fh2>\u003Cp>如果拿 Claude 跟 \u003Ca href=\"https:\u002F\u002Fopenai.com\" target=\"_blank\" rel=\"noopener\">OpenAI\u003C\u002Fa> 或 \u003Ca href=\"https:\u002F\u002Fai.google\u002F\" target=\"_blank\" rel=\"noopener\">Google\u003C\u002Fa> 比，Anthropic 的文件風格真的比較偏工程師。它不是一直講願景，而是把 token、價格、alias、版本、端點全部列出來。\u003C\u002Fp>\u003Cp>這種寫法有個好處。你可以直接做成本試算。假設一個產品每月吃 20M input token 和 5M output token，Opus 4.7 的成本大概是 $125 + $125 = $250。Sonnet 4.6 會是 $60 + $75 = $135。Haiku 4.5 則是 $20 + $25 = $45。差距很直接。\u003C\u002Fp>\u003Cp>所以問題不是「哪個最好」。問題是「哪個夠用」。很多團隊會想先上最強模型，但真的上線後，錢和延遲會教你做人。Anthropic 這次把選擇邊界畫得很清楚，算是幫你少踩幾個坑。\u003C\u002Fp>\u003Cul>\u003Cli>Opus 4.7 適合複雜推理、長流程 agent、程式碼生成\u003C\u002Fli>\u003Cli>Sonnet 4.6 適合一般產品功能與成本控制\u003C\u002Fli>\u003Cli>Haiku 4.5 適合高併發、低延遲任務\u003C\u002Fli>\u003Cli>價格差距很大，輸出成本最容易拉開總帳單\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這份文件透露的，其實是 Anthropic 的產品策略\u003C\u002Fh2>\u003Cp>我覺得這次更新不只是文件整理。Anthropic 正在把 Claude 做成一套分層產品，而不是一顆萬用模型。這種策略很像雲端服務的產品線，先分清楚高階、中階、輕量，再讓開發者自己配。\u003C\u002Fp>\u003Cp>這也代表 API 使用方式會更成熟。你不再只看 benchmark 分數。你還要看 context window、max output、alias、cloud endpoint、版本鎖定。對做 AI 產品的人來說，這些細節比模型宣傳影片重要太多。\u003C\u002Fp>\u003Cp>如果你現在已經在用 Claude，下一步很簡單。先看你的工作負載。再看你到底是卡推理、卡成本，還是卡延遲。很多時候，Sonnet 就夠了。真的需要重推理和長輸出，再切到 Opus。\u003C\u002Fp>\u003Ch2>結論：先別迷信最大模型，先算帳\u003C\u002Fh2>\u003Cp>這次 Claude 文件更新，最有價值的地方不是 Opus 4.7 三個字，而是 Anthropic 把規則講清楚了。對開發者來說，這能少掉很多猜測，也能少掉很多錯配。\u003C\u002Fp>\u003Cp>我的建議很直接。先把你的 API 成本、token 用量、輸出長度列出來。再決定要不要升到 Opus 4.7。很多團隊最後會發現，真正適合的不是最貴那顆，而是最剛好的那顆。\u003C\u002Fp>","Anthropic 把 Claude 文件重心放到 Opus 4.7，整理 1M token 上下文、128k 輸出上限、定價與跨雲模型 ID。","platform.claude.com","https:\u002F\u002Fplatform.claude.com\u002Fdocs\u002Fen\u002Fabout-claude\u002Fmodels\u002Foverview",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1779023645425-mkdf.png","model-release","zh","7314af2f-8c90-47cf-98a7-4ee0407c61d5",[17,18,19,20,21,22,23,24,25,26],"Claude","Anthropic","Claude API","Opus 4.7","Sonnet 4.6","Haiku 4.5","Model ID","Bedrock","Vertex AI","AI 模型選擇",[28,29,30],"Anthropic 把 Opus 4.7 放到文件最上層，主打複雜推理與 agentic coding。","1M token 上下文已成為 Opus 與 Sonnet 的共同門檻，差別更集中在輸出上限與價格。","模型 ID、alias、雲端端點與版本鎖定規則更清楚，對 production 團隊很實用。",3,"2026-05-17T13:13:36.325561+00:00","2026-05-17T13:13:36.278+00:00","0ccb5d2e-69f1-4354-a3e0-cb370221cd95",{"tags":36,"relatedLang":47,"relatedPosts":51},[37,39,41,43,45],{"name":19,"slug":38},"claude-api",{"name":20,"slug":40},"opus-47",{"name":18,"slug":42},"anthropic",{"name":17,"slug":44},"claude",{"name":21,"slug":46},"sonnet-46",{"id":15,"slug":48,"title":49,"language":50},"claude-api-model-guide-new-top-tier-en","Claude API model guide gets a new top tier","en",[52,58,64,70,76,82],{"id":53,"slug":54,"title":55,"cover_image":56,"image_url":56,"created_at":57,"category":13},"1985ce38-03c6-4968-96fa-b751553bbef3","why-claude-opus-48-is-not-the-big-story-zh","為什麼 Claude Opus 4.8 不是大新聞","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780531367297-nrfs.png","2026-06-04T00:02:24.633987+00:00",{"id":59,"slug":60,"title":61,"cover_image":62,"image_url":62,"created_at":63,"category":13},"8810b91a-9aa2-4cd6-a58b-18fad5897423","devin-booker-sedona-mcdonalds-shoe-launch-zh","Booker把Sedona麥當勞變鞋款發表場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780510686292-fm1k.png","2026-06-03T18:17:31.966783+00:00",{"id":65,"slug":66,"title":67,"cover_image":68,"image_url":68,"created_at":69,"category":13},"d4d7e664-cc7f-4211-a733-b7c111b86bd6","best-open-source-llms-2026-ranked-zh","2026 最佳開源 LLM 排名","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780396385004-yyka.png","2026-06-02T10:32:37.264398+00:00",{"id":71,"slug":72,"title":73,"cover_image":74,"image_url":74,"created_at":75,"category":13},"06774dfe-08eb-4a53-a8f7-36389b462c2b","llama-3-1-70b-specs-benchmarks-deployment-zh","Llama 3.1 70B：規格與部署","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780395481064-5yri.png","2026-06-02T10:17:33.072306+00:00",{"id":77,"slug":78,"title":79,"cover_image":80,"image_url":80,"created_at":81,"category":13},"e8ee6f00-cf62-41e6-83b7-92ce148fe46e","kill-bill-whole-bloody-affair-4k-blu-ray-zh","《追殺比爾：血腥全集》4K 藍光上市","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780364908972-15qn.png","2026-06-02T01:48:00.707278+00:00",{"id":83,"slug":84,"title":85,"cover_image":86,"image_url":86,"created_at":87,"category":13},"893178f1-7aba-4a0c-a3cf-1812c9d3283e","almalinux-10-2-9-8-new-stacks-zh","AlmaLinux 10.2 與 9.8 更新了什麼","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780291073047-7bxy.png","2026-06-01T05:17:27.940241+00:00",[89,94,99,104,109,114,119,124,129,134],{"id":90,"slug":91,"title":92,"created_at":93},"58b64033-7eb6-49b9-9aab-01cf8ae1b2f2","nvidia-rubin-six-chips-one-ai-supercomputer-zh","NVIDIA Rubin 把六顆晶片塞進 AI 機櫃","2026-03-26T07:18:45.861277+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"0dcc2c61-c2a6-480d-adb8-dd225fc68914","march-2026-ai-model-news-what-mattered-zh","2026 年 3 月 AI 模型新聞重點","2026-03-26T07:32:08.386348+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"214ab08b-5ce5-4b5c-8b72-47619d8675dd","why-small-models-are-winning-on-device-ai-zh","小模型為何吃下裝置端 AI","2026-03-26T07:36:30.488966+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"785624b2-0355-4b82-adc3-de5e45eecd88","midjourney-v8-faster-images-higher-costs-zh","Midjourney V8 變快了，也變貴了","2026-03-26T07:52:03.562971+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"cda76b92-d209-4134-86c1-a60f5bc7b128","xiaomi-mimo-trio-agents-robots-voice-zh","小米 MiMo 三模型瞄準代理、機器人與語音","2026-03-28T03:05:08.779489+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"9e1044b4-946d-47fe-9e2a-c2ee032e1164","xiaomi-mimo-v2-pro-1t-moe-agents-zh","小米 MiMo-V2-Pro 登場：1T MoE 模型","2026-03-28T03:06:19.002353+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"c4b6186f-bd84-4598-997e-c6e31d543c0d","cursor-composer-2-agentic-coding-model-zh","Cursor Composer 2 走向代理式寫碼","2026-03-28T03:13:06.422716+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"e112e76f-ec3b-408f-810e-e93ae21a888a","apple-siri-gemini-distilled-models-zh","Apple Siri 牽手 Gemini 的真相","2026-03-29T04:52:57.886544+00:00",{"id":130,"slug":131,"title":132,"created_at":133},"c679b51f-194a-463b-87fc-7695256ff752","mimo-v2-pro-vs-omni-vs-flash-2026-zh","MiMo V2 Pro、Omni、Flash 怎麼選","2026-04-02T01:18:43.576128+00:00",{"id":135,"slug":136,"title":137,"created_at":138},"3b988fd7-6749-4f01-ba25-c0ad7486dc31","z-ai-glm-5v-turbo-design2code-claude-zh","GLM-5V-Turbo 在 Design2Code 贏了…","2026-04-02T04:03:36.31741+00:00"]