[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-kimi-k27-whats-new-and-how-to-run-it-zh":3,"article-related-kimi-k27-whats-new-and-how-to-run-it-zh":30,"series-model-release-2495916e-6109-4bd6-a948-f118ebd948ca":75},{"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":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"2495916e-6109-4bd6-a948-f118ebd948ca","kimi-k27-whats-new-and-how-to-run-it-zh","Kimi K2.7 上線與驗證清單","\u003Cp data-speakable=\"summary\">這篇教你確認 Kimi K2.7 模型 ID、做 A\u002FB 測試，並把它接進現有 \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> 工作流。\u003C\u002Fp>\u003Cp>這篇給正在評估 \u003Ca href=\"\u002Ftag\u002Fmoonshot-ai\">Moonshot AI\u003C\u002Fa> Kimi K2.7 的開發者看，尤其是已經在用 Kimi K2.6、OpenRouter，或長駐 agent 架構的人。照做完，你會拿到一份可直接執行的驗證流程，知道它是否適合你的\u003Ca href=\"\u002Ftag\u002F長上下文\">長上下文\u003C\u002Fa>、中文與多步工具任務。\u003C\u002Fp>\u003Cp>你也會得到一個可落地的切換方法，從模型確認、測試集建立，到接入既有應用與觀察結果，都能一路照著做，不需要重寫整個系統。\u003C\u002Fp>\u003Ch2>開始之前\u003C\u002Fh2>\u003Cul>\u003Cli>OpenRouter 帳號與 API 存取權\u003C\u002Fli>\u003Cli>有效的 OpenRouter API key\u003C\u002Fli>\u003Cli>Node 20+ 或 Python 3.11+ 本機環境\u003C\u002Fli>\u003Cli>可依模型名稱切換的 agent app 或 playground\u003C\u002Fli>\u003Cli>5 到 10 筆真實提示詞、文件或工單樣本\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fopenclawlaunch.com\u002Fblog\u002Fkimi-k2-7-release\">OpenClaw Launch 的 Kimi K2.7 發布文\u003C\u002Fa>\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fopenrouter.ai\u002Fmodels\">OpenRouter model list\u003C\u002Fa>\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FOpenRouterTeam\u002Fopenrouter-sdk\">OpenRouter SDK\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Step 1: 確認 K2.7 模型 ID\u003C\u002Fh2>\u003Cp>目的：先抓到 OpenRouter 上的精確模型名稱，避免把舊 ID 或錯誤別名寫進正式設定。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781791368143-oe0k.png\" alt=\"Kimi K2.7 上線與驗證清單\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>打開 OpenRouter model list，找到 Kimi K2.7 項目，記下它的 ID、context length 與定價欄位。後續所有測試與部署都以這個 live listing 為準，不要只看文章標題。\u003C\u002Fp>\u003Cpre>\u003Ccode>curl https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1\u002Fmodels | jq '.data[] | select(.name | test(\"Kimi.*K2.7\"; \"i\")) | {id, name, context_length, pricing}'\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該看到一筆可直接貼進設定檔的 Kimi K2.7 模型 ID。\u003C\u002Fp>\u003Ch2>Step 2: 建立 A\u002FB 測試題庫\u003C\u002Fh2>\u003Cp>目的：做出一組固定題目，讓 K2.7 能和你現有模型在同一批任務上公平比較。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781791366926-d9n0.png\" alt=\"Kimi K2.7 上線與驗證清單\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>挑 5 到 10 題，覆蓋你的真實工作：長文件摘要、中文改寫、程式碼修改、多步除錯，以及一題多模態或圖片輸入。每一題都要固定內容，確保不同模型看到的是同一份輸入。\u003C\u002Fp>\u003Cp>把題目存成純文字或 JSON，方便 agent 自動重播。如果你手上已有 production traces，直接拿來用，比自造測試樣本更有參考價值。\u003C\u002Fp>\u003Cp>驗收：你\u003Ca href=\"\u002Fnews\u002Fgovernment-can-pull-unsafe-ai-models-offline-zh\">應該有\u003C\u002Fa>一份可重跑、可比對的題庫檔案，能對不同模型產生一致輸入。\u003C\u002Fp>\u003Ch2>Step 3: 切換 agent 到 Kimi K2.7\u003C\u002Fh2>\u003Cp>目的：用最小改動把既有 agent 指到 K2.7，先驗證它能在真實流程中正常回應。\u003C\u002Fp>\u003Cp>如果你用 \u003Ca href=\"\u002Ftag\u002Fopenclaw\">OpenClaw\u003C\u002Fa> 或 Hermes Agent，就在模型下拉選單選 K2.7，再填入 OpenRouter key。若你是直接呼叫 API，只要改 request 裡的 model 欄位，其他參數先維持不變。\u003C\u002Fp>\u003Cpre>\u003Ccode>import OpenAI from \"openai\";\n\nconst client = new OpenAI({\n  baseURL: \"https:\u002F\u002Fopenrouter.ai\u002Fapi\u002Fv1\",\n  apiKey: process.env.OPENROUTER_API_KEY,\n});\n\nconst response = await client.chat.completions.create({\n  model: \"moonshot\u002Fkimi-k2-7\",\n  messages: [\n    { role: \"system\", content: \"You are a coding agent.\" },\n    { role: \"user\", content: \"Review this repo and propose a fix.\" }\n  ]\n});\n\nconsole.log(response.choices[0].message.content);\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>驗收：你應該收到一個有效 completion，而且路由、工具與儲存流程都沒有被迫改寫。\u003C\u002Fp>\u003Ch2>Step 4: 跑逐題對照測試\u003C\u002Fh2>\u003Cp>目的：用你自己的資料判斷 K2.7 是否真的優於基準模型，而不是只看「新模型」印象。\u003C\u002Fp>\u003Cp>把每一題分別送進 K2.7 與基準模型，逐項評分 correctness、usefulness、tool-call quality，若是程式任務再看 edit distance。長上下文題目要特別檢查前段資訊是否被保留，以及模型能否一路遵守指令到最後。\u003C\u002Fp>\u003Cp>先保持條件一致：temperature 相同、max tokens 相同、工具配置相同。若結果怪異，再一次只改一個變數，方便定位差異來源。\u003C\u002Fp>\u003Cp>驗收：你應該能明確指出哪個模型在摘要、中文寫作與 \u003Ca href=\"\u002Ftag\u002Fagentic-coding\">agentic coding\u003C\u002Fa> 上更適合你的工作流。\u003C\u002Fp>\u003Ch2>Step 5: 把勝出模型放進常駐 agent\u003C\u002Fh2>\u003Cp>目的：把測試結果\u003Ca href=\"\u002Fnews\u002Fweb3-turns-platform-control-into-user-ownership-zh\">變成\u003C\u002Fa>可持續運作的預設設定，讓日常流量直接吃到最佳模型。\u003C\u002Fp>\u003Cp>當你確認勝出者後，把預設模型改到 agent config，並保留基準模型作為 fallback。若平台支援下拉選單，讓 K2.7 成為長文件與中文任務的預設，快速低成本請求則維持較小模型。\u003C\u002Fp>\u003Cp>\u003Ca href=\"\u002Fnews\u002Flinux-kernel-7-1-fred-ntfs-amd-fixes-zh\">上線\u003C\u002Fa>時採漸進式釋出，觀察 latency、cost 與回答品質在真實流量下的表現。如果 K2.7 品質更好但速度較慢，就把它保留給真正值得的任務。\u003C\u002Fp>\u003Cp>驗收：你應該看到 live requests 穩定走 K2.7，且使用者端沒有中斷，fallback 也可隨時切回。\u003C\u002Fp>\u003Ch2>常見錯誤\u003C\u002Fh2>\u003Cul>\u003Cli>只看文章，不看 live model list。修法：上線前先從 OpenRouter 複製最新 model ID。\u003C\u002Fli>\u003Cli>不同模型卻用不同題目或參數。修法：固定測試集、temperature 與工具設定，確保比較公平。\u003C\u002Fli>\u003Cli>所有任務都無差別改用 K2.7。修法：只把它留給長上下文、中文或多步 agent 任務。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>接下來可以看什麼\u003C\u002Fh2>\u003Cp>完成 K2.7 驗證後，可以用同一套 A\u002FB 方法擴到其他模型，接著整理出每一層任務該用哪個模型，讓團隊之後切換時有明確依據。\u003C\u002Fp>","這篇教你確認 Kimi K2.7 模型 ID、做 A\u002FB 測試，並把它接進現有 agent 工作流。","openclawlaunch.com","https:\u002F\u002Fopenclawlaunch.com\u002Fblog\u002Fkimi-k2-7-release",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781791368143-oe0k.png","model-release","zh","952ab890-dacd-429b-93d2-3821a5dc00bc",[17,18,19,20,21],"Kimi K2.7","OpenRouter","agent workflow","A\u002FB testing","Node 20+",[23,24,25],"先用 OpenRouter live model list 確認 K2.7 的精確 ID，再寫入設定檔。","用固定題庫做 A\u002FB 測試，才能公平比較 K2.7 和基準模型。","只在長上下文、中文與 agentic 任務上優先導入 K2.7，保留 fallback。",0,"2026-06-18T14:02:24.876682+00:00","2026-06-18T14:02:24.867+00:00","0ccb5d2e-69f1-4354-a3e0-cb370221cd95",{"tags":31,"relatedLang":34,"relatedPosts":38},[32],{"name":18,"slug":33},"openrouter",{"id":15,"slug":35,"title":36,"language":37},"kimi-k27-whats-new-and-how-to-run-it-en","Kimi K2.7: What Changed and How to Run It","en",[39,45,51,57,63,69],{"id":40,"slug":41,"title":42,"cover_image":43,"image_url":43,"created_at":44,"category":13},"a419fc45-bd6c-4ce2-a2ef-2a0467f6c02d","kimi-k27-code-highspeed-mode-skips-benchmarks-zh","Kimi K2.7-Code 主打快，但證據還不夠","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781795889545-t5hx.png","2026-06-18T15:17:40.944644+00:00",{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":13},"98c49728-7b3d-4f35-8835-4bfeddb0aa12","linux-kernel-7-1-fred-ntfs-amd-fixes-zh","Linux 7.1 上線：FRED、NTFS、AMD 一次補齊","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781787771307-ptmn.png","2026-06-18T13:02:24.621368+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":13},"9ff13e71-8310-491e-8564-75de9520a3ea","fable-5-drew-rare-praise-ai-voices-zh","Fable 5 為何引發 AI 圈關注","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781748177844-sx21.png","2026-06-18T02:02:30.664296+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":13},"1a90c708-48c0-4e89-a0b5-8f8b6d4b05e9","devin-pricing-june-2026-plans-limits-zh","Devin 2026 年 6 月定價拆解","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781735572958-8i40.png","2026-06-17T22:32:27.692488+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":13},"976800ba-7c59-4890-a17c-866a751f4f61","self-host-minimax-m3-gpu-cloud-zh","MiniMax M3 自架 GPU 雲部署分析","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781716686607-r9gm.png","2026-06-17T17:17:35.332244+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":13},"1d05b1ff-9ced-4fde-bc0d-f21e4775c8c8","apple-gemini-backed-ai-still-its-own-thing-zh","Apple 的 Gemini 血統，還是 Apple 的 AI","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781706792148-e040.png","2026-06-17T14:32:44.523843+00:00",[76,81,86,91,96,101,106,111,116,121],{"id":77,"slug":78,"title":79,"created_at":80},"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":82,"slug":83,"title":84,"created_at":85},"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":87,"slug":88,"title":89,"created_at":90},"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":92,"slug":93,"title":94,"created_at":95},"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":97,"slug":98,"title":99,"created_at":100},"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":102,"slug":103,"title":104,"created_at":105},"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":107,"slug":108,"title":109,"created_at":110},"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":112,"slug":113,"title":114,"created_at":115},"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":117,"slug":118,"title":119,"created_at":120},"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":122,"slug":123,"title":124,"created_at":125},"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"]