[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-why-kimi-k25-changes-open-source-agent-race-zh":3,"article-related-why-kimi-k25-changes-open-source-agent-race-zh":29,"series-model-release-0735b585-9553-4864-8a7e-9149aacacbf7":73},{"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":11},"0735b585-9553-4864-8a7e-9149aacacbf7","why-kimi-k25-changes-open-source-agent-race-zh","為什麼 Kimi K2.5 會改寫開源 Agent 競賽","\u003Cp data-speakable=\"summary\">Kimi K2.5 把多模態推理、工具調用和多代理協作綁在一起，讓開源 A\u003Ca href=\"\u002Fnews\u002Faws-bedrock-knowledge-bases-rag-zh\">ge\u003C\u002Fa>nt 從會聊天變成能做事。\u003C\u002Fp>\u003Cp>Kimi K2.5 不是又一次\u003Ca href=\"\u002Fnews\u002Fmulti-fidelity-models-composite-mechanics-zh\">模型\u003C\u002Fa>發表，而是把開源 AI 從「能回答」推向「能執行」的明確訊號。它的重點不在單次問答更漂亮，而在能否跨視覺、文字與工具鏈完成真實工作流程，這才是開源 \u003Ca href=\"\u002Ftag\u002Fagent\">Agent\u003C\u002Fa> 競賽真正的分水嶺。\u003C\u002Fp>\u003Cp>從公開資訊看，K2.5 是原生多模態模型，訓練資料量約 15T 的圖文混合 \u003Ca href=\"\u002Ftag\u002Ftoken\">token\u003C\u002Fa>，並且搭配可自我調度的 agent swarm，最多能協調 100 個子代理、處理高達 1,500 次工具調用。這已經不是「偶爾會呼叫工具的聊天模型」，而是一套以執行為中心的系統。\u003C\u002Fp>\u003Ch2>第一個論點\u003C\u002Fh2>\u003Cp>\u003Ca href=\"\u002Ftag\u002F開源模型\">開源模型\u003C\u002Fa>過去太常把文字能力當成全部，但真實工作不是文字測驗。Kimi K2.5 直接把這個前提推翻：它主打同時具備 coding 與 vision 能力，且視覺與文字預訓練不是互相犧牲，而是疊加強化。官方展示的網站影片重建案例就很有代表性，因為它不是玩具級標註，而是「看懂畫面再產出可用程式」的工作流。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777965039267-f8wa.png\" alt=\"為什麼 Kimi K2.5 會改寫開源 Agent 競賽\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>數據也支持這個方向。K2.5 在不使用工具時，文字分數 31.5、圖像分數 21.3；接上工具後，文字提升到 51.8、圖像提升到 39.8。這個差距說明一件事：模型的價值不只在內部推理，而在能否接上環境、搜尋資訊、拆解步驟並完成動作。對工程團隊來說，這才是能否落地的門檻。\u003C\u002Fp>\u003Ch2>第二個論點\u003C\u002Fh2>\u003Cp>這次最重要的突破，不是單一模型變大，而是它如何組織工作。官方說法指出，K2.5 能\u003Ca href=\"\u002Fnews\u002Fspeckv-adaptive-speculative-decoding-gamma-zh\">自動\u003C\u002Fa>建立並協調 agent swarm，不需要預先定義子代理或固定流程，還能透過平行執行把 runtime 最多縮短 4.5 倍。這直接打到應用 AI 的核心痛點：單一模型可以推理，但很難有效放大自己的努力。\u003C\u002Fp>\u003Cp>對產品與開發團隊而言，這比榜單名次更重要。瀏覽器任務、程式重構、文件審閱、試算表整理，這些工作不只是「回答問題」，而是要拆分、並行、交叉驗證。Kimi.com 與 app 提供的四種模式，包括 Agent Swarm beta，顯示 Moonshot 想賣的不是一個聊天框，而是一個工作平台。這是基礎設施思維，不是 demo 思維。\u003C\u002Fp>\u003Ch2>反方可能怎麼說\u003C\u002Fh2>\u003Cp>質疑者的說法其實很合理：benchmark 很容易被包裝，工具調用常常只是把模型弱點外包給外部系統，多代理協作也常因協調成本而失控。再加上官方提到為了避免資料外洩，Hugging Face 版本被限制存取，這會讓人懷疑可重現性、評測潔淨度，以及這些提升是否只在自家環境有效。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777965045168-yl20.png\" alt=\"為什麼 Kimi K2.5 會改寫開源 Agent 競賽\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>這個批評有力，但不足以推翻 K2.5 的意義。因為這次的價值不在抽象的「純模型分數」，而在整合後的實際能力。若模型在能看、能查、能拆工時明顯更強，那不是作弊，而是產品定義。當然，限制也存在：團隊不能把它當成宇宙通用智慧，而要在自己的資料、工具與流程裡實測。\u003C\u002Fp>\u003Cp>換句話說，K2.5 的爭議不是它有沒有完美，而是它把 AI 的競爭焦點從「誰的單點能力最高」改成「誰能把推理變成可重複的執行」。這個轉向本身就足夠重要，因為大多數企業需要的不是一個更會說話的模型，而是一個能穩定縮短流程時間的系統。\u003C\u002Fp>\u003Ch2>你能做什麼\u003C\u002Fh2>\u003Cp>如果你是工程師，應該拿 K2.5 去測那些最容易暴露淺層推理的任務：視覺除錯、前端還原、文件抽取、試算表自動化、多步驟程式修改。如果你是 PM 或創辦人，不要再問「它是不是最強」，而要問「它能不能把原本要三個工具、兩個人完成的流程，壓成一個可控的工作流」。K2.5 的訊號很明確：下一輪 AI 競爭，不只比模型品質，更比跨文字、視覺與工具的協同執行能力。\u003C\u002Fp>","Kimi K2.5 不是單純更強的模型，而是把多模態推理、工具調用與多代理協作綁成一個可執行的開源系統。","www.kimi.com","https:\u002F\u002Fwww.kimi.com\u002Fblog\u002Fkimi-k2-5",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777965039267-f8wa.png","model-release","zh","f54893ef-5e99-4326-9120-88752dbc8b4d",[17,18,19,20,21],"Kimi K2.5","開源 Agent","多模態推理","工具調用","多代理協作",[23,24,25],"Kimi K2.5 的關鍵不是更會聊天，而是更會執行。","多模態能力加上工具與 agent swarm，才是開源 AI 的新門檻。","真正該測的是工作流效率，不是單一榜單分數。",3,"2026-05-05T07:10:23.992168+00:00","2026-05-05T07:10:23.892+00:00",{"tags":30,"relatedLang":11,"relatedPosts":36},[31,33,34,35],{"name":18,"slug":32},"開源-agent",{"name":19,"slug":19},{"name":20,"slug":20},{"name":21,"slug":21},[37,43,49,55,61,67],{"id":38,"slug":39,"title":40,"cover_image":41,"image_url":41,"created_at":42,"category":13},"003f59ba-8d5a-40cb-8e6b-0b51898bc537","midjourney-21-second-video-model-closed-ai-wrong-deal-zh","為什麼 Midjourney 的 21 秒影片模型證明封閉式 AI 是錯的交易","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780648396657-oa18.png","2026-06-05T08:32:37.400103+00:00",{"id":44,"slug":45,"title":46,"cover_image":47,"image_url":47,"created_at":48,"category":13},"b5889da8-fa42-44ed-89a7-3347655b388d","microsoft-seven-ai-models-openai-anthropic-build-2026-zh","7 款 Microsoft AI 模型登場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780642975404-f8mr.png","2026-06-05T07:02:23.607092+00:00",{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"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":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"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":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"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":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"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",[74,79,84,89,94,99,104,109,114,119],{"id":75,"slug":76,"title":77,"created_at":78},"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":80,"slug":81,"title":82,"created_at":83},"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":85,"slug":86,"title":87,"created_at":88},"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":90,"slug":91,"title":92,"created_at":93},"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":95,"slug":96,"title":97,"created_at":98},"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":100,"slug":101,"title":102,"created_at":103},"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":105,"slug":106,"title":107,"created_at":108},"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":110,"slug":111,"title":112,"created_at":113},"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":115,"slug":116,"title":117,"created_at":118},"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":120,"slug":121,"title":122,"created_at":123},"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"]