[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-cursor-kimi-k25-disclosure-miss-explained-zh":3,"tags-cursor-kimi-k25-disclosure-miss-explained-zh":33,"related-lang-cursor-kimi-k25-disclosure-miss-explained-zh":43,"related-posts-cursor-kimi-k25-disclosure-miss-explained-zh":47,"series-tools-4e504c31-5fea-4a7a-be2a-125425772378":84},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":10,"image_url":11,"keywords":12,"language":21,"translated_content":10,"views":22,"is_premium":23,"created_at":24,"updated_at":24,"cover_image":11,"published_at":25,"rewrite_status":26,"rewrite_error":10,"rewritten_from_id":27,"slug":28,"category":29,"related_article_id":30,"status":31,"google_indexed_at":32,"x_posted_at":10,"tweet_text":10,"title_rewritten_at":10,"title_original":10,"key_takeaways":10,"topic_cluster_id":10,"embedding":10,"is_canonical_seed":23},"4e504c31-5fea-4a7a-be2a-125425772378","Cursor 漏報 Kimi K2.5，問題在哪","\u003Cp>Cursor 先說 Composer 2 跑出 61.7% 的 Terminal-Bench 2.0。還說價格比 Claude Opus 4.6 低很多。結果有人抓 API 流量，直接看到模型 ID 指向 \u003Ca href=\"https:\u002F\u002Fwww.moonshot.cn\u002Fen\" target=\"_blank\" rel=\"noopener\">Moonshot AI\u003C\u002Fa> 的 \u003Ca href=\"https:\u002F\u002Fwww.moonshot.cn\u002Fen\u002Fkimi\" target=\"_blank\" rel=\"noopener\">Kimi K2.5\u003C\u002Fa>。說真的，這種落差很難裝沒事。\u003C\u002Fp>\u003Cp>這件事不是少寫一行 credit 而已。它牽涉授權、供應商信任，還有一個很現實的問題：你把程式碼丟進 AI 助理時，底層到底是哪個模型在看？如果答案不清楚，企業法務和資安團隊就會開始皺眉頭。\u003C\u002Fp>\u003Ch2>Cursor 說了什麼，API 又揭露了什麼\u003C\u002Fh2>\u003Cp>Cursor 在 3 月 19 日推出 Composer 2。官方文案寫得很像自家練出來的模型。它提到 continued pretraining、reinforcement learning，還有 frontier-level coding intelligence。這些詞都沒錯，數字也有兜住。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775157170056-atpp.png\" alt=\"Cursor 漏報 Kimi K2.5，問題在哪\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>但問題出在出身。開發者 Fynn 看 Cursor 的 API 流量後，抓到一串很像線索的模型 ID：\u003Ccode>kimi-k2p5-rl-0317-s515-fast\u003C\u002Fcode>。這個命名很直白，幾乎就是把底牌攤開。它指向 \u003Ca href=\"\u002Fnews\u002Fkimi-k2-5-pricing-2026-plans-api-costs-zh\">Kimi\u003C\u002Fa> K2.5、RL 微調、3 月 17 日訓練點，以及 fast 服務配置。\u003C\u002Fp>\u003Cp>後來 Cursor 也承認，當初應該先講清楚 \u003Ca href=\"\u002Fnews\u002Fkimi-k25-brings-vision-code-and-swarm-agents-zh\">Kimi\u003C\u002Fa> base model。Aman Sanger 直接說這是漏掉了。講白了，這就把故事從「我們自己做的」改成「我們在開放模型上再訓練」。兩者差很多，尤其在商業溝通上。\u003C\u002Fp>\u003Cul>\u003Cli>發表時間：2026 年 3 月 19 日\u003C\u002Fli>\u003Cli>Terminal-Bench 2.0：61.7%\u003C\u002Fli>\u003Cli>價格主張：大約是 Claude Opus 4.6 的十分之一\u003C\u002Fli>\u003Cli>模型 ID：\u003Ccode>accounts\u002Fanysphere\u002Fmodels\u002Fkimi-k2p5-rl-0317-s515-fast\u003C\u002Fcode>\u003C\u002Fli>\u003Cli>公開討論：相關貼文衝到百萬級瀏覽\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>授權條款為什麼會卡住\u003C\u002Fh2>\u003Cp>Kimi K2.5 用的是改過的 MIT 授權。這種授權平常很寬鬆，商用也通常沒問題。可是它有一條很硬的條件。月活超過 1 億，或月營收超過 2,000 萬美元，產品介面就得明顯標示「Kimi K2.5」。\u003C\u002Fp>\u003Cp>Cursor 的年經常性收入傳聞已超過 20 億美元。換算下來，大約每月 1.67 億美元。這個數字遠遠超過門檻。也就是說，這不是「要不要客氣一下」的層級，而是直接碰到授權義務。\u003C\u002Fp>\u003Cp>後續還有一段小插曲。Moonshot AI 的員工先在網路上指出違規，之後又刪文。Moonshot 官方帳號後來說，這是透過 \u003Ca href=\"https:\u002F\u002Ffireworks.ai\" target=\"_blank\" rel=\"noopener\">Fireworks AI\u003C\u002Fa> 的授權商業合作。這可能解掉一部分法律問題，但不代表發表時沒漏報。\u003C\u002Fp>\u003Cblockquote>“It was a miss to not mention the Kimi base in our blog from the start.” — Aman Sanger, Cursor co-founder\u003C\u002Fblockquote>\u003Cp>這句話很直接，也很有用。因為它承認的不是技術錯誤，而是揭露不完整。對工程團隊來說，這種差別很重要。你可以接受供應商用別人的基座，但你不能接受供應商把這件事講得像沒發生過。\u003C\u002Fp>\u003Ch2>為什麼開發團隊應該在意\u003C\u002Fh2>\u003Cp>AI 產品現在都是疊起來的。行銷頁面看到的模型，只是最上層。下面可能還有開放權重基座、供應商微調、另一家推理服務商，最後才是你看到的 UI。每一層都有自己的 license 和資料路徑。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775157174056-0cpe.png\" alt=\"Cursor 漏報 Kimi K2.5，問題在哪\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這對資安、採購、合規都很重要。你如果有 GDPR、HIPAA，或資料落地要求，就不能只聽一句「我們有合規」。你要知道 prompt 送去哪裡、誰在處理、log 存在哪裡。這些都不是細節，是基本盤。\u003C\u002Fp>\u003Cp>信任也是同一件事。若供應商把 base model、訓練方式和推理位置講得很清楚，你至少能查證。若它把產品包裝成「自研」，卻不提底層模型，那你拿到的其實是行銷話術，不是技術事實。\u003C\u002Fp>\u003Cul>\u003Cli>先問 base model 是哪個\u003C\u002Fli>\u003Cli>再問推理服務商是誰\u003C\u002Fli>\u003Cli>再問資料和 log 存哪裡\u003C\u002Fli>\u003Cli>最後問授權要求什麼標示\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Kimi K2.5 跟西方選項怎麼比\u003C\u002Fh2>\u003Cp>Kimi K2.5 不是小玩具。它是 1 兆參數的 mixture-of-ex\u003Ca href=\"\u002Fnews\u002Fopera-neon-adds-mcp-support-for-ai-clients-zh\">per\u003C\u002Fa>ts 模型，32 億 active parameters，context window 也有 256,000 tokens。這種規格很適合長程式碼編輯、整個 repo 上下文，還有 agent 型工作流。\u003C\u002Fp>\u003Cp>西方開源陣營這陣子沒有那麼整齊。\u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fllama\u002F\" target=\"_blank\" rel=\"noopener\">Meta Llama\u003C\u002Fa> 4 Scout 和 Maverick 有出，但市場反應沒有很多人預期的強。Llama 4 Behemoth 也還沒公開發布日期。對想找強力開放基座的團隊來說，空窗期還在。\u003C\u002Fp>\u003Cp>這也說明一件事。AI 模型市場早就不是單純的地緣站隊。Cursor 是美國公司，估值 500 億美元，結果底層用了北京公司的模型。這不是意外，而是性能導向下的正常選擇。講白了，誰好用就先上誰。\u003C\u002Fp>\u003Cul>\u003Cli>Kimi K2.5：1 兆參數，32 億 active，256k context\u003C\u002Fli>\u003Cli>Llama 4 Behemoth：延後，沒有公開日期\u003C\u002Fli>\u003Cli>Cursor Composer 2：Kimi K2.5 基座再加 RL 訓練\u003C\u002Fli>\u003Cli>Cursor 自述：大約 25% 算力來自 base model，75% 來自自家訓練\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這件事放在產業脈絡裡看\u003C\u002Fh2>\u003Cp>現在很多 AI 工具都在比誰更像「原生產品」。但實際上，模型供應鏈越來越像雲端服務堆疊。前端是產品，後端是 API，中間還有模型商、推理商、快取、評測和監控。你看到的品牌，不一定就是訓練那個品牌。\u003C\u002Fp>\u003Cp>這種模式在 coding assistant 特別常見。因為工程師最在意的是延遲、上下文長度、錯誤率和價格。只要模型夠強，很多團隊不會先問它是不是自研。可是一旦進到企業採購，問題就會回來。法務和資安不看 benchmark，他們看合同和資料流。\u003C\u002Fp>\u003Cp>所以這次事件真正的價值，是把「模型來源」這件事拉到台面上。以前大家只看分數。現在你還得看 provenance。這不是潔癖，是採購基本功。\u003C\u002Fp>\u003Ch2>開發者現在該怎麼做\u003C\u002Fh2>\u003Cp>如果你的團隊有在用 AI 寫 code，就把模型來源當成 dependency metadata。不要只看介面好不好用。先問清楚 base model、推理商、授權條件和資料保存方式。這些東西最好寫進採購清單。\u003C\u002Fp>\u003Cp>供應商也該學軟體圈老方法。package 要列版本，container 要列 base image，AI 產品也該列模型卡和訓練說明。這不是裝模作樣。這是讓客戶知道自己到底買了什麼。\u003C\u002Fp>\u003Cp>我自己的看法很簡單。下一次選 AI coding tool，先別看宣傳圖。先問一句：底層模型是誰？如果答案模糊，先當黑盒處理。等它把來源講清楚，再談要不要把 production code 交出去。\u003C\u002Fp>\u003Cp>如果你想把這件事放進團隊流程，最實際的做法是：把模型名稱、授權條款、資料保留政策，全部寫進採購審查表。這樣至少下次出事時，不會只能靠社群抓包。\u003C\u002Fp>","Cursor 的 Composer 2 宣稱自研，卻被發現底層其實是 Kimi K2.5。這次漏報不只影響信任，也牽涉授權、資料流向和企業採購判斷。","dev.to","https:\u002F\u002Fdev.to\u002Fharsh2644\u002Fcursor-used-kimi-k25-a-chinese-ai-model-without-disclosure-why-every-developer-should-care-15h6",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775157170056-atpp.png",[13,14,15,16,17,18,19,20],"Cursor","Kimi K2.5","AI coding assistant","模型揭露","授權條款","資料治理","Moonshot AI","Claude Opus 4.6","zh",1,false,"2026-04-02T19:12:34.237234+00:00","2026-04-02T19:12:34.097+00:00","done","4e40de6b-5828-4ef3-a8ee-f79f23de23e7","cursor-kimi-k25-disclosure-miss-explained-zh","tools","9d46441a-edbe-4866-9950-a1c7f229a693","published","2026-04-08T09:00:49.507+00:00",[34,36,37,39,41,42],{"name":13,"slug":35},"cursor",{"name":18,"slug":18},{"name":15,"slug":38},"ai-coding-assistant",{"name":19,"slug":40},"moonshot-ai",{"name":16,"slug":16},{"name":17,"slug":17},{"id":30,"slug":44,"title":45,"language":46},"cursor-kimi-k25-disclosure-miss-explained-en","Cursor’s Kimi K2.5 Disclosure Miss, Explained","en",[48,54,60,66,72,78],{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":29},"d058a76f-6548-4135-8970-f3a97f255446","why-gemini-api-pricing-is-cheaper-than-it-looks-zh","為什麼 Gemini API 定價其實比看起來更便宜","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778869845081-j4m7.png","2026-05-15T18:30:25.797639+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":29},"68e4be16-dc38-4524-a6ea-5ebe22a6c4fb","why-vidhub-huiyuan-hutong-bushi-quan-shebei-tongyong-zh","為什麼 VidHub 會員互通不是「買一次全設備通用」","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778789450987-advz.png","2026-05-14T20:10:24.048988+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":29},"7a1e174f-746b-4e82-a0e3-b2475ab39747","why-buns-zig-to-rust-experiment-is-right-zh","為什麼 Bun 的 Zig-to-Rust 實驗是對的","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778767879127-5dna.png","2026-05-14T14:10:26.886397+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":29},"e742fc73-5a65-4db3-ad17-88c99262ceb7","why-openai-api-pricing-is-product-strategy-zh","為什麼 OpenAI API 定價是產品策略，不是註腳","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778749859485-chvz.png","2026-05-14T09:10:26.003818+00:00",{"id":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"category":29},"c757c5d8-eda9-45dc-9020-4b002f4d6237","why-claude-code-prompt-design-beats-ide-copilots-zh","為什麼 Claude Code 的提示設計贏過 IDE Copilot","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778742645084-dao9.png","2026-05-14T07:10:29.371901+00:00",{"id":79,"slug":80,"title":81,"cover_image":82,"image_url":82,"created_at":83,"category":29},"4adef3ab-9f07-4970-91cf-77b8b581b348","why-databricks-model-serving-is-right-default-zh","為什麼 Databricks Model Serving 是生產推論的正確預設","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778692245329-a2wt.png","2026-05-13T17:10:30.659153+00:00",[85,90,95,100,105,110,115,120,125,130],{"id":86,"slug":87,"title":88,"created_at":89},"de769291-4574-4c46-a76d-772bd99e6ec9","googles-biggest-gemini-launches-in-2026-zh","Google 2026 最大 Gemini 盤點","2026-03-26T07:26:39.21072+00:00",{"id":91,"slug":92,"title":93,"created_at":94},"855cd52f-6fab-46cc-a7c1-42195e8a0de4","surepath-real-time-mcp-policy-controls-zh","SurePath 推出即時 MCP 政策控管","2026-03-26T07:57:40.77233+00:00",{"id":96,"slug":97,"title":98,"created_at":99},"9b19ab54-edef-4dbd-9ce4-a51e4bae4ebb","mcp-in-2026-the-ai-tool-layer-teams-use-zh","2026 年 MCP：團隊真的在用的 AI 工具層","2026-03-26T08:01:46.589694+00:00",{"id":101,"slug":102,"title":103,"created_at":104},"af9c46c3-7a28-410b-9f04-32b3de30a68c","prompting-in-2026-what-actually-works-zh","2026 提示工程，真正有用的是什麼","2026-03-26T08:08:12.453028+00:00",{"id":106,"slug":107,"title":108,"created_at":109},"05553086-6ed0-4758-81fd-6cab24b575e0","garry-tan-open-sources-claude-code-toolkit-zh","Garry Tan 開源 Claude Code 工具包","2026-03-26T08:26:20.068737+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"042a73a2-18a2-433d-9e8f-9802b9559aac","github-ai-projects-to-watch-in-2026-zh","2026 必看 20 個 GitHub AI 專案","2026-03-26T08:28:09.619964+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"a5f94120-ac0d-4483-9a8b-63590071ac6a","claude-code-vs-cursor-2026-zh","Claude Code 與 Cursor 深度對比：202…","2026-03-26T13:27:14.279193+00:00",{"id":121,"slug":122,"title":123,"created_at":124},"0975afa1-e0c7-4130-a20d-d890eaed995e","practical-github-guide-learning-ml-2026-zh","2026 機器學習入門 GitHub 實用指南","2026-03-27T01:16:49.712576+00:00",{"id":126,"slug":127,"title":128,"created_at":129},"bfdb467a-290f-4a80-b3a9-6f081afb6dff","aiml-2026-student-ai-ml-lab-repo-review-zh","AIML-2026：像課綱的學生實驗 Repo","2026-03-27T01:21:51.467798+00:00",{"id":131,"slug":132,"title":133,"created_at":134},"80cabc3e-09fc-4ff5-8f07-b8d68f5ae545","ai-trending-github-repos-and-research-feeds-zh","AI Trending：把 AI 資源收成一張表","2026-03-27T01:31:35.262183+00:00"]