[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"tag-長上下文":3},{"tag":4,"articles":10,"peer_article_count":149},{"id":5,"name":6,"slug":6,"article_count":7,"description_zh":8,"description_en":9},"87866137-c0da-4bc0-a289-aca4b3445de2","長上下文",7,"長上下文指模型能在同一次推理中保留更多文件、程式碼、對話與工具輸出，從 128K、256K 到百萬級 token 都是重點。它影響長文件分析、跨檔案編輯、代理式工作流與記憶壓縮策略，也直接牽動成本、延遲與幻覺風險。","Long context refers to models that can keep far more text in a single run, from 128K and 256K windows to million-token APIs. It matters for codebases, long documents, agent workflows, memory compression, and the tradeoffs between cost, latency, and reliability.",[11,20,28,35,42,50,57,64,71,78,86,93,100,107,114,121,128,135,142],{"id":12,"slug":13,"title":14,"summary":15,"category":16,"image_url":17,"cover_image":17,"language":18,"created_at":19},"ca1e6960-10e7-4fa7-949f-c5991c99fc7e","kimi-k26-open-source-coding-agentic-ai-benchmarks-zh","Kimi K2.6 登頂程式與代理式 AI 基準","Moonshot AI 在 2026-06-26 發布 Kimi K2.6，主打 262,144 token 長上下文、300 子代理與 4,000 步協作，並在多項程式與代理式基準拿下高分。","model-release","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782739078329-qvne.png","zh","2026-06-29T13:17:26.530857+00:00",{"id":21,"slug":22,"title":23,"summary":24,"category":25,"image_url":26,"cover_image":26,"language":18,"created_at":27},"e1c96c63-93c0-4cc0-8e69-26cbd0655457","turboquant-cuts-llm-memory-use-without-retraining-zh","TurboQuant 讓長上下文推理更省記憶體","5 項重點看懂 TurboQuant 如何在不重訓下壓縮 KV cache，將記憶體用量最多降 6×，並在長上下文推理中提升吞吐。","industry","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782710266206-izlz.png","2026-06-29T05:17:22.332572+00:00",{"id":29,"slug":30,"title":31,"summary":32,"category":16,"image_url":33,"cover_image":33,"language":18,"created_at":34},"aaf20836-acd9-42ef-b247-481d82e6a26d","minimax-m3-open-weight-frontier-models-matter-zh","MiniMax M3 證明開放權重前沿模型已經重要","MiniMax M3 顯示開放權重模型已能在程式碼、代理、長上下文與多模態上，和前沿閉源模型正面競爭。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782388970243-o5tn.png","2026-06-25T12:02:23.509174+00:00",{"id":36,"slug":37,"title":38,"summary":39,"category":16,"image_url":40,"cover_image":40,"language":18,"created_at":41},"8b0b6a07-b173-42ab-883a-77d720808276","kimi-long-context-models-moonshot-ai-zh","Kimi 的長上下文一路加大","Moonshot AI 的 Kimi 從長上下文聊天機器人，走到 agent 與 1T 參數模型。Kimi K2.5 在 2026 年 1 月登場，也把產品線推到更複雜的階段。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782231484525-xqgo.png","2026-06-23T16:17:38.02879+00:00",{"id":43,"slug":44,"title":45,"summary":46,"category":47,"image_url":48,"cover_image":48,"language":18,"created_at":49},"7171fed6-f304-4f46-9efe-f691ea304b65","randomized-yarn-long-context-reasoning-zh","Randomized YaRN 讓長上下文更穩","Randomized YaRN 透過隨機化位置編碼與長度課程，讓只看過短上下文訓練的 LLM，更能推廣到 16K 到 128K 的長推理窗口。","research","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782195475543-rsm6.png","2026-06-23T06:17:32.36653+00:00",{"id":51,"slug":52,"title":53,"summary":54,"category":25,"image_url":55,"cover_image":55,"language":18,"created_at":56},"f298de46-73ff-425c-9941-f58b4e43adce","xiaomi-mimo-code-beats-claude-code-long-tasks-zh","小米 MiMo Code 挑戰 Claude Code","小米推出開源 MiMo Code，主打 200 步以上的長任務編碼表現，並同步釋出 MiMo Auto 與 MiMo-V2.5 系列參數、價格與上下文數據。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781559171329-8ofv.png","2026-06-15T21:32:19.550119+00:00",{"id":58,"slug":59,"title":60,"summary":61,"category":16,"image_url":62,"cover_image":62,"language":18,"created_at":63},"b15b0887-bd5b-43e6-ac42-23939d0f4e92","google-gemini-35-pro-june-2m-token-launch-zh","Gemini 3.5 Pro 6月登場，2M Token 夠猛","Google 傳出要在 6 月推出 Gemini 3.5 Pro，主打 2M Token 上下文。這代表長文件、程式碼庫和多輪分析會更好處理，但實際表現還是要看價格、速度和穩定性。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1781204585839-bdsh.png","2026-06-11T19:02:36.371587+00:00",{"id":65,"slug":66,"title":67,"summary":68,"category":16,"image_url":69,"cover_image":69,"language":18,"created_at":70},"66ce4542-3c93-4a0c-ab52-5e6f90a36212","minimax-m3-kai-fang-quan-zhong-xie-cheng-shi-reng-neng-ying-zh","MiniMax M3 證明開放權重在寫程式上仍能贏","MiniMax M3 證明開放權重模型不只可以追上前沿，還能在寫程式、長上下文與成本控制上形成優勢。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780968786191-lele.png","2026-06-09T01:32:30.829528+00:00",{"id":72,"slug":73,"title":74,"summary":75,"category":16,"image_url":76,"cover_image":76,"language":18,"created_at":77},"409fc126-8ed2-42e3-bec3-9d114c4aca23","why-minimax-m3-matters-long-context-model-zh","為什麼 MiniMax M3 比又一個長上下文模型更重要","MiniMax M3 的重要性不在於它又把上下文做大，而在於它把長上下文、多模態與代理控制綁成一個可用系統。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780755468369-c0ia.png","2026-06-06T14:17:20.522361+00:00",{"id":79,"slug":80,"title":81,"summary":82,"category":83,"image_url":84,"cover_image":84,"language":18,"created_at":85},"bef47dbc-b0b4-439e-bae9-abe9473a321c","wei-shen-me-tether-ba-ben-di-ai-ji-yi-tui-jin-ri-chang-zhuan-zh","為什麼 Tether 把本地 AI 記憶推進日常裝置是對的","TurboQuant 的價值不在於更快，而在於把長上下文 AI 從資料中心拉回手機、筆電與邊緣裝置，讓本地 AI 真正可用。","tools","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780542170805-opi6.png","2026-06-04T03:02:19.599329+00:00",{"id":87,"slug":88,"title":89,"summary":90,"category":16,"image_url":91,"cover_image":91,"language":18,"created_at":92},"06774dfe-08eb-4a53-a8f7-36389b462c2b","llama-3-1-70b-specs-benchmarks-deployment-zh","Llama 3.1 70B：規格與部署","Meta 的 Llama 3.1 70B 仍是 128K 長上下文的自架文字模型，適合內部聊天、RAG 與 API 編排，重點在成本控制與部署自主性。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780395481064-5yri.png","2026-06-02T10:17:33.072306+00:00",{"id":94,"slug":95,"title":96,"summary":97,"category":83,"image_url":98,"cover_image":98,"language":18,"created_at":99},"2f8b506f-91a9-4d0c-9171-303301c4d23a","why-claude-code-should-use-deepseek-v4-for-1m-context-zh","為什麼 Claude Code 應該用 DeepSeek v4 來處理 1M …","Claude Code 在長上下文程式工作上，應優先路由到 DeepSeek v4，因為 1M context 比品牌偏好更能決定實際產出。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777867842186-4psy.png","2026-05-04T04:10:18.556889+00:00",{"id":101,"slug":102,"title":103,"summary":104,"category":16,"image_url":105,"cover_image":105,"language":18,"created_at":106},"b875d3ed-f892-43a8-a51e-920729e85b1e","gpt-5-4-benchmarks-2026-scores-rankings-zh","GPT-5.4 知識測驗拿 97.6 分","GPT-5.4 在 BenchLM 知識與理解拿到 97.6 分，總榜暫列第 2，還有 1.05M token 長上下文。這篇拆解它適合哪些工作、和其他模型怎麼比。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1776082194973-cyii.png","2026-04-13T12:09:40.301446+00:00",{"id":108,"slug":109,"title":110,"summary":111,"category":83,"image_url":112,"cover_image":112,"language":18,"created_at":113},"99c0866d-50f9-4a93-a282-b092f9d298df","claude-code-compaction-context-management-zh","Claude Code壓縮機制怎麼省上下文","Claude Code 用多層壓縮處理長對話上下文，避免 200K 到 1M token 被文件、Shell 輸出和編輯記錄吃光。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775629324817-9lcw.png","2026-04-08T06:21:46.65172+00:00",{"id":115,"slug":116,"title":117,"summary":118,"category":16,"image_url":119,"cover_image":119,"language":18,"created_at":120},"fad499f8-512b-4d92-8110-7a4aaac4801f","grok-41-xai-quieter-upgrade-matters-zh","Grok 4.1 低調升級，卻很有料","xAI 的 Grok 4.1 把幻覺率從 12.09% 降到 4.22%，還加入 Fast 與 Thinking 兩種模式，支援 256k context 與 2M token API，對開發者很實際。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775175345966-349k.png","2026-04-03T00:15:29.860687+00:00",{"id":122,"slug":123,"title":124,"summary":125,"category":16,"image_url":126,"cover_image":126,"language":18,"created_at":127},"f0fb0635-5207-4fc5-b913-a4ab205ebb66","grok-420-xai-flagship-model-explained-zh","Grok 4.20 怎麼看","xAI 的 Grok 4.20 主打 200 萬 token 長上下文、多代理推理與 API 價格。這篇拆解它的定位、規格、競品差異與開發者該注意的坑。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775175176314-zyny.png","2026-04-03T00:12:37.401835+00:00",{"id":129,"slug":130,"title":131,"summary":132,"category":25,"image_url":133,"cover_image":133,"language":18,"created_at":134},"ff021fab-7330-4e01-8187-ca099f7c31f4","claude-vs-chatgpt-copilot-gemini-enterprise-2026-zh","Claude、ChatGPT、Copilot、Gemini…","Claude 擅長長上下文與程式工作；ChatGPT、Copilot、Gemini 則靠分發、整合與工作流吃香。企業 2026 年該怎麼選，重點不是誰最強，而是誰最適合你的資料、流程與控管。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775153757367-xo7r.png","2026-04-02T18:15:40.452457+00:00",{"id":136,"slug":137,"title":138,"summary":139,"category":16,"image_url":140,"cover_image":140,"language":18,"created_at":141},"5a3c6417-77a9-4526-bee5-c355979576f2","gemini-3-1-pro-googles-top-model-in-numbers-zh","Gemini 3.1 Pro 數字看真實力","Gemini 3.1 Pro 以 77.1% ARC-AGI-2、94.3% GPQA Diamond、1M token 上下文登場，價格仍維持 Gemini 3。這次重點不是噱頭，而是長文檔、程式碼與 agent 工作流的實戰成本。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775153580311-vv9w.png","2026-04-02T18:12:41.777858+00:00",{"id":143,"slug":144,"title":145,"summary":146,"category":47,"image_url":147,"cover_image":147,"language":18,"created_at":148},"9d1ed0f2-aace-46ce-9b0a-0c0d8655e8e8","turboquant-wont-fix-memory-crunch-zh","TurboQuant 解不了記憶體荒","Google 的 TurboQuant 可把 KV-cache 記憶體用量降到 6 倍，但更長上下文、更多 agent 與更高吞吐，可能把 DRAM 和 NAND 需求繼續往上推。","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775132150405-6fvw.png","2026-04-02T12:15:31.810812+00:00",0]