[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-moonshot-ai-kimi-models-growth-zh":3,"article-related-moonshot-ai-kimi-models-growth-zh":33,"series-industry-b69d875b-d141-46a1-87ee-1b173e88a6e7":78},{"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":25,"views":29,"created_at":30,"published_at":31,"topic_cluster_id":32},"b69d875b-d141-46a1-87ee-1b173e88a6e7","moonshot-ai-kimi-models-growth-zh","Kimi 模型正在快速變大","\u003Cp data-speakable=\"summary\">\u003Ca href=\"\u002Ftag\u002Fmoonshot-ai\">Moonshot AI\u003C\u002Fa> 的 \u003Ca href=\"\u002Fnews\u002Fkimi-k27-code-highspeed-mode-skips-benchmarks-zh\">Kimi\u003C\u002Fa> 系列已經做到 1 兆參數、256K 上下文，Kimi K2.5 也加上視覺能力。\u003C\u002Fp>\u003Cp>\u003Ca href=\"https:\u002F\u002Fwww.moonshot.ai\" target=\"_blank\" rel=\"noopener\">Moonshot AI\u003C\u002Fa> 從 2023 年 3 月在北京成立，到現在只花不到 3 年。它已經變成中國 AI 圈最常被盯著看的公司之一。講白了，這家公司走得很快，\u003Ca href=\"\u002Fnews\u002Fkimi-k27-whats-new-and-how-to-run-it-zh\">Kimi\u003C\u002Fa> 模型也跟著長很快。\u003C\u002Fp>\u003Cp>這次的重點很直接。\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fmoonshotai\u002FKimi-K2.5\" target=\"_blank\" rel=\"noopener\">Kimi K2.5\u003C\u002Fa> 把視覺能力拉進來。前一版 \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fmoonshotai\u002FKimi-K2\" target=\"_blank\" rel=\"noopener\">Kimi K2\u003C\u002Fa> 已經有 1 兆參數。現在再往上加，路線很明確，就是往更大、更長、更全能走。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>數字\u003C\u002Fth>\u003Cth>意義\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>公司成立時間\u003C\u002Ftd>\u003Ctd>2023 年 3 月\u003C\u002Ftd>\u003Ctd>成長速度很快\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Kimi K2 參數\u003C\u002Ftd>\u003Ctd>1 兆\u003C\u002Ftd>\u003Ctd>模型規模已進入超大級別\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>上下文長度\u003C\u002Ftd>\u003Ctd>256K\u003C\u002Ftd>\u003Ctd>可處理更長文件與對話\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Kimi K2.5\u003C\u002Ftd>\u003Ctd>加入視覺能力\u003C\u002Ftd>\u003Ctd>開始走多模態路線\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>Kimi 為什麼一直往上加\u003C\u002Fh2>\u003Cp>先講結論。這不是單純比誰參數大。Moonshot AI 在做的是把\u003Ca href=\"\u002Ftag\u002F長上下文\">長上下文\u003C\u002Fa>、工具使用、推理和視覺，全部塞進同一條產品線。對開發者來說，這比只看 \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> 分數更實際。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781798565646-ltv8.png\" alt=\"Kimi 模型正在快速變大\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>你可能會想問，1 兆參數到底有多誇張。講白了，這代表訓練、推理、部署都更吃資源。模型一大，伺服器壓力就上來，成本也不會客氣。這也是為什麼很多團隊會先做小模型，再慢慢往上堆。\u003C\u002Fp>\u003Cp>Kimi 的策略比較兇。它不是只做聊天機器人，而是把長文件整理、程式輔助、圖片理解一起包進去。這種做法很像在搶「工作入口」，不是只搶一個聊天框。\u003C\u002Fp>\u003Cul>\u003Cli>1 兆參數，代表模型容量很大。\u003C\u002Fli>\u003Cli>256K 上下文，適合長文件。\u003C\u002Fli>\u003Cli>K2.5 加視覺，開始走多模態。\u003C\u002Fli>\u003Cli>產品定位更像通用 AI 工具。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這條路線和其他模型差在哪\u003C\u002Fh2>\u003Cp>拿 \u003Ca href=\"https:\u002F\u002Fopenai.com\u002Fgpt-4\" target=\"_blank\" rel=\"noopener\">GPT-4\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fdeepmind.google\u002Ftechnologies\u002Fgemini\u002F\" target=\"_blank\" rel=\"noopener\">Gemini\u003C\u002Fa> 來看，大家都在往長上下文和多模態走。差別在節奏。Moonshot AI 這邊的動作很密，幾乎是一路往上推。\u003C\u002Fp>\u003Cp>對比之下，很多模型會把重點放在推理或成本。Kimi 比較像是先把能力堆滿，再看怎麼落地。這種打法有風險，因為模型越大，延遲和價格越難壓。但它也有好處，就是容易吸引重度使用者。\u003C\u002Fp>\u003Cp>如果你是開發者，會很在意三件事。第一是 \u003Ca href=\"\u002Ftag\u002Fapi\">API\u003C\u002Fa> 穩不穩。第二是上下文到底能不能真的吃滿。第三是視覺能力是不是只是 demo，而是真的能進工作流。這三個問題，比「模型很大」更重要。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Ca href=\"https:\u002F\u002Fopenai.com\u002Fgpt-4\" target=\"_blank\" rel=\"noopener\">GPT-4\u003C\u002Fa> 強在生態和產品整合。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fwww.anthropic.com\u002Fclaude\" target=\"_blank\" rel=\"noopener\">Claude\u003C\u002Fa> 很會處理長文和寫作。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fdeepmind.google\u002Ftechnologies\u002Fgemini\u002F\" target=\"_blank\" rel=\"noopener\">Gemini\u003C\u002Fa> 在多模態整合很積極。\u003C\u002Fli>\u003Cli>\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fmoonshotai\u002FKimi-K2.5\" target=\"_blank\" rel=\"noopener\">Kimi K2.5\u003C\u002Fa> 則是把規模和視覺一起推上去。\u003C\u002Fli>\u003C\u002Ful>\u003Cblockquote>\"There is no free lunch.\" — Leslie Valiant\u003C\u002Fblockquote>\u003Cp>這句話很適合拿來看大模型。模型變大，不代表一切都變好。它只代表你把問題往別的地方搬。訓練更貴，推理更慢，部署更麻煩，這些都要有人買單。\u003C\u002Fp>\u003Cp>Moonshot AI 走這條路，等於是先把上限拉高。接下來就看它能不能把成本壓下來，還能維持體驗。這才是最硬的地方。\u003C\u002Fp>\u003Ch2>數字背後是什麼競爭\u003C\u002Fh2>\u003Cp>中國 AI 市場的節奏很快。\u003Ca href=\"https:\u002F\u002Fwww.baidu.com\" target=\"_blank\" rel=\"noopener\">Baidu\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.alibabacloud.com\" target=\"_blank\" rel=\"noopener\">Alibaba Cloud\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.tencentcloud.com\" target=\"_blank\" rel=\"noopener\">Tencent Cloud\u003C\u002Fa> 都有自己的模型路線。Moonshot AI 要在這裡站住腳，不能只靠聲量。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781798559502-blcx.png\" alt=\"Kimi 模型正在快速變大\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>256K 上下文是很實用的賣點。對法務、研究、程式碼庫、產品規格書，這種長文件場景很有用。很多團隊不是缺聊天功能，是缺能\u003Ca href=\"\u002Fnews\u002Flinux-kernel-7-1-fred-ntfs-amd-fixes-zh\">一次\u003C\u002Fa>吃下整包資料的模型。\u003C\u002Fp>\u003Cp>但大模型競爭很現實。你要看價格、延遲、穩定度、錯誤率，還有 API 文件寫得好不好。少一項都會影響導入。說真的，很多模型死在文件太爛，不是死在能力不夠。\u003C\u002Fp>\u003Cul>\u003Cli>參數大，不等於最好用。\u003C\u002Fli>\u003Cli>長上下文對企業場景很重要。\u003C\u002Fli>\u003Cli>視覺能力能擴大使用場景。\u003C\u002Fli>\u003Cli>API 和價格才是導入關鍵。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>這家公司現在站在哪裡\u003C\u002Fh2>\u003Cp>Moonshot AI 的成長很像中國新創圈的縮影。先用一個能打的產品切進市場，再靠模型升級維持聲量。這套路不新，但執行得快，就能搶到注意力。\u003C\u002Fp>\u003Cp>對\u003Ca href=\"\u002Ftag\u002F台灣開發者\">台灣開發者\u003C\u002Fa>來說，這件事也有參考價值。你要做 AI 產品，不一定一開始就拚最大模型。先看資料來源、工作流程、成本結構，再決定要不要接大模型。很多時候，模型只是最後一塊拼圖。\u003C\u002Fp>\u003Cp>如果 Kimi K2.5 真的把視覺、長文、推理整合得夠順，它會逼其他廠商加速更新。反過來說，如果體驗跟不上，參數再大也只是數字遊戲。市場很現實，使用者也不吃這套。\u003C\u002Fp>\u003Ch2>接下來該看什麼\u003C\u002Fh2>\u003Cp>我覺得下一步不是看它又多了幾個參數，而是看它的 API、價格和延遲。這三個數字一出來，大家就知道它是不是能進真實專案。\u003C\u002Fp>\u003Cp>如果你在評估模型，先拿你的文件、圖片、程式碼去測。不要只看 demo。真正的答案，通常藏在你的資料裡，不在發表頁面上。\u003C\u002Fp>\u003Cp>Moonshot AI 已經把 Kimi 推到很大。接下來最值得追的，不是它會不會再長，而是它能不能把這些能力變成穩定、便宜、可用的服務。\u003C\u002Fp>","Moonshot AI 的 Kimi 系列已走到 1 兆參數、256K 上下文，Kimi K2.5 也加入視覺能力，速度很快。","en.wikipedia.org","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FMoonshot_AI",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781798565646-ltv8.png","industry","zh","c0e27fca-3077-4e46-abc7-4f388feb20e5",[17,18,19,20,21,22,23,24],"Moonshot AI","Kimi","Kimi K2.5","1兆參數","256K上下文","多模態","大模型","AI",[26,27,28],"Kimi 系列已經走到 1 兆參數和 256K 上下文。","Kimi K2.5 加入視覺能力，路線往多模態前進。","真正要看的不是參數，而是 API、價格和延遲。",0,"2026-06-18T16:02:18.797393+00:00","2026-06-18T16:02:18.787+00:00","caa87b65-9bbc-46fe-bba8-4f4158dd2d8b",{"tags":34,"relatedLang":37,"relatedPosts":41},[35],{"name":17,"slug":36},"moonshot-ai",{"id":15,"slug":38,"title":39,"language":40},"moonshot-ai-kimi-models-growth-en","Moonshot AI’s Kimi models are getting bigger fast","en",[42,48,54,60,66,72],{"id":43,"slug":44,"title":45,"cover_image":46,"image_url":46,"created_at":47,"category":13},"f2177583-b1dd-4de8-8adf-6754d7c8f9c2","hermes-agent-learning-memory-gateways-zh","Hermes Agent 的 5 個關鍵能力","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781867872053-6r9l.png","2026-06-19T11:17:23.11904+00:00",{"id":49,"slug":50,"title":51,"cover_image":52,"image_url":52,"created_at":53,"category":13},"39cf67a3-ce5a-4a4c-9322-a42a6e6476c8","rust-built-different-update-server-admins-zh","Rust 6 月更新讓伺服器管理員頭痛","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781851683502-ql4f.png","2026-06-19T06:47:36.493495+00:00",{"id":55,"slug":56,"title":57,"cover_image":58,"image_url":58,"created_at":59,"category":13},"7088e08a-2624-4ee5-bc20-136f59f3a627","openai-ipo-prep-policy-hiring-play-zh","OpenAI IPO 前先把政策變成招募","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781842703362-p8sd.png","2026-06-19T04:17:56.622121+00:00",{"id":61,"slug":62,"title":63,"cover_image":64,"image_url":64,"created_at":65,"category":13},"8e153a8b-a625-4797-a8cb-b7fbc7273810","openai-right-to-hire-dean-ball-policy-power-zh","OpenAI 延攬 Dean Ball 是對的：政策權力已是 AI 核心戰場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781841764947-t4hd.png","2026-06-19T04:02:18.251835+00:00",{"id":67,"slug":68,"title":69,"cover_image":70,"image_url":70,"created_at":71,"category":13},"7d221962-c85f-4776-af7f-cedd921d67fb","rust-rolling-release-model-right-tradeoff-zh","Rust 的滾動式支援，才是團隊真正該接受的取捨","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781834562661-8eed.png","2026-06-19T02:02:18.140322+00:00",{"id":73,"slug":74,"title":75,"cover_image":76,"image_url":76,"created_at":77,"category":13},"5d19eff9-8a4f-4ca4-ac8a-e793ebadf7f4","cloudflare-director-vote-pressure-ai-push-zh","Cloudflare 面臨董事投票壓力","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781830969389-5tea.png","2026-06-19T01:02:25.790674+00:00",[79,84,89,94,99,104,109,114,119,124],{"id":80,"slug":81,"title":82,"created_at":83},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":85,"slug":86,"title":87,"created_at":88},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":90,"slug":91,"title":92,"created_at":93},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":125,"slug":126,"title":127,"created_at":128},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]