[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-why-xiaomi-mimo-v25-pro-changes-coding-agents-zh":3,"article-related-why-xiaomi-mimo-v25-pro-changes-coding-agents-zh":31,"series-model-release-d63e9d93-e613-4bbf-8135-9599fde11d08":84},{"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":23,"views":27,"created_at":28,"published_at":29,"topic_cluster_id":30},"d63e9d93-e613-4bbf-8135-9599fde11d08","why-xiaomi-mimo-v25-pro-changes-coding-agents-zh","為什麼 Xiaomi 的 MiMo-V2.5-Pro 改變的是 Coding …","\u003Cp data-speakable=\"summary\">MiMo-V2.5-Pro 是為長時間、重工具呼叫的程式工作設計，不是為聊天而生。\u003C\u002Fp>\u003Cp>Xiaomi 的 MiMo-V2.5-Pro 不是又一次榜單秀肌肉，而是清楚告訴市場：下一輪 AI 競爭比的是耐力，不是嘴皮子。官方給出的案例很直接，這個模型能在 4.3 小時內做出完整編譯器，經過 672 次 tool c\u003Ca href=\"\u002Fnews\u002Fgala-games-web3-gaming-2026-zh\">al\u003C\u002Fa>ls 後把測試覆蓋率推到 100%，而且在 coding 分數上接近 \u003Ca href=\"\u002Ftag\u002Fclaude\">Claude\u003C\u002Fa> Opus 4.6，同時還少用 40% 到 60% 的 tokens。\u003C\u002Fp>\u003Ch2>第一個論點\u003C\u002Fh2>\u003Cp>真正改變市場的，不是它能不能寫一段程式，而是它能不能把一個任務一路做完。Xiaomi 展示的編譯器案例最有說服力：一開始只有 59% coverage，途中還因為重構把自己改壞，最後能自己發現錯誤並回到 233\u002F233 hidden tests 全過。這不是 autocomplete，而是帶著狀態往前推進的工程工作。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778689858139-v38e.png\" alt=\"為什麼 Xiaomi 的 MiMo-V2.5-Pro 改變的是 Coding …\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>第二個例子更能說明問題。Xiaomi 說它用約 1,870 次 tool calls，花 11.5 小時，無監督完成了一個約 8,000 行 code 的桌面影片編輯器。這種任務的關鍵不是單次回答品質，而是模型能不能記住前文、修正前一步、再接著往下做。對 coding age\u003Ca href=\"\u002Fnews\u002Fanthropic-1-8b-akamai-compute-deal-zh\">nt\u003C\u002Fa> 來說，這才是新的勝負手。\u003C\u002Fp>\u003Ch2>第二個論點\u003C\u002Fh2>\u003Cp>token efficiency 已經不是加分題，而是產品能不能落地的核心條件。Xiaomi 直接宣稱，MiMo-V2.5-Pro 比 Claude Opus 4.6 和 \u003Ca href=\"\u002Ftag\u002Fgemini\">Gemini\u003C\u002Fa> 3.1 Pro 少用 40% 到 60% 的 tokens。對長時間 \u003Ca href=\"\u002Ftag\u002Fagent\">agent\u003C\u002Fa> 工作來說，這代表成本、延遲和失敗面都一起下降。少一半 token，不只是省錢，還意味著同樣的工作可以更穩定地重跑、回滾和擴大部署。\u003C\u002Fp>\u003Cp>上下文長度也支持這個判斷。主版本可處理到 100 萬 tokens，基礎版不經再訓練也有 256,000 tokens；Xiaomi 還說，local 與 global attention 的組合把記憶體用量壓低將近 7 倍，parallel token prediction 則把輸出速度拉高 3 倍。這些數字對聊天產品未必是決勝點，但對要在 agent loop 裡跑幾個小時的系統，直接決定可不可以上線。\u003C\u002Fp>\u003Ch2>反方可能怎麼說\u003C\u002Fh2>\u003Cp>最強的反對意見其實很合理：\u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> 贏，不等於真實產品場景也贏。模型可以在內部編譯器任務上表現漂亮，卻在混亂的 repository、破碎的 API、模糊的需求和會反悔的人類審查面前失手。長上下文也不是萬靈丹，MiMo-V2-Pro 先前在 \u003Ca href=\"\u002Ftag\u002Fopenai\">OpenAI\u003C\u002Fa> 的 GraphWalks benchmark 上，據稱在 100 萬 token 長度下拿到 0 分，這提醒我們長距推理仍然會崩。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778689872679-vxzf.png\" alt=\"為什麼 Xiaomi 的 MiMo-V2.5-Pro 改變的是 Coding …\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>另一個質疑是，硬體公司做的 open-weight release 常常看起來很強，實際上卻難部署、難微調、難整合進工程流程。MiMo-V2.5-Pro 的內部測試、token 節省和 staged post-training，都還是控制良好的展示，不是帶著 leg\u003Ca href=\"\u002Fnews\u002Fwhy-lace-20-matters-more-than-cardanos-next-hard-fork-zh\">ac\u003C\u002Fa>y dependencies、權限問題和人類干擾的真實 codebase。\u003C\u002Fp>\u003Cp>這些批評成立，但不足以推翻結論。MiMo-V2.5-Pro 的重點不是宣告軟體工程已被解決，而是說明評價標準已經變了。現在更有價值的是能處理多小時任務、保住上下文、並在自己犯錯後修正回來的模型。就算部分 demo 經過挑選，方向仍然清楚，Xiaomi 公布的數字也足以證明：coding agent 的競爭焦點，已經不是聊天品質。\u003C\u002Fp>\u003Ch2>你能做什麼\u003C\u002Fh2>\u003Cp>如果你是工程師或 PM，現在該做的不是追最新聊天模型，而是把工作流改成長任務導向：把流程拆成可觀測階段、記錄每次 tool call、量 token burn，並替模型中途失準設計回復機制。若你是創辦人，請直接拿幾個模型做長時間 agent 測試，重點看誰能在幾小時內穩定完成任務，而不是誰在單輪對話裡最會回答。這一輪的贏家不是最會聊天的模型，而是最會把事做完的模型。\u003C\u002Fp>","MiMo-V2.5-Pro 的重點不在聊天能力，而在長時間、重工具呼叫的 coding agent 工作；它代表 AI 競爭焦點正從會說話，轉向能把任務做完。","www.neura.market","https:\u002F\u002Fwww.neura.market\u002Fnews\u002Fxiaomi-mimo-v2-5-pro-open-model-compiler-coding-benchmarks",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778689858139-v38e.png","model-release","zh","1d5fc6b1-a87f-48ae-89ee-e5f0da86eb2d",[17,18,19,20,21,22],"Xiaomi","MiMo-V2.5-Pro","coding agent","open weights","long-context","tool use",[24,25,26],"MiMo-V2.5-Pro 的價值在長任務與工具呼叫，不在聊天感。","token 效率與上下文長度，正在成為 agent 產品能否落地的關鍵。","開放權重策略讓這類模型更容易進入真實開發流程。",4,"2026-05-13T16:30:27.893951+00:00","2026-05-13T16:30:27.759+00:00","0ccb5d2e-69f1-4354-a3e0-cb370221cd95",{"tags":32,"relatedLang":43,"relatedPosts":47},[33,35,37,39,41],{"name":18,"slug":34},"mimo-v25-pro",{"name":17,"slug":36},"xiaomi",{"name":19,"slug":38},"coding-agent",{"name":40,"slug":21},"long context",{"name":20,"slug":42},"open-weights",{"id":15,"slug":44,"title":45,"language":46},"why-xiaomi-mimo-v25-pro-changes-coding-agents-en","Why Xiaomi’s MiMo-V2.5-Pro Changes Coding Agents More Than Chatbots","en",[48,54,60,66,72,78],{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":13},"b5926931-ce20-4b9d-8814-a3c960187209","what-we-know-about-gpt-56-release-date-zh","GPT-5.6 何時發布？目前線索整理","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780574585815-dzo7.png","2026-06-04T12:02:35.122398+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"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":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"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":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"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":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"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":79,"slug":80,"title":81,"cover_image":82,"image_url":82,"created_at":83,"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",[85,90,95,100,105,110,115,120,125,130],{"id":86,"slug":87,"title":88,"created_at":89},"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":91,"slug":92,"title":93,"created_at":94},"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":96,"slug":97,"title":98,"created_at":99},"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":101,"slug":102,"title":103,"created_at":104},"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":106,"slug":107,"title":108,"created_at":109},"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":111,"slug":112,"title":113,"created_at":114},"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":116,"slug":117,"title":118,"created_at":119},"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":121,"slug":122,"title":123,"created_at":124},"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":126,"slug":127,"title":128,"created_at":129},"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":131,"slug":132,"title":133,"created_at":134},"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"]