[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-open-source-llm-comparison-2026-zh":3,"article-related-open-source-llm-comparison-2026-zh":30,"series-model-release-710ff4cc-d333-4bd8-b50a-e5522d430161":81},{"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":11,"views":27,"created_at":28,"published_at":29,"topic_cluster_id":11},"710ff4cc-d333-4bd8-b50a-e5522d430161","open-source-llm-comparison-2026-zh","2026 開源 LLM 誰領先","\u003Cp>2026 年的開源 LLM，不再只是玩具。\u003Ca href=\"https:\u002F\u002Fcomputingforgeeks.com\u002Fopen-source-llm-comparison\u002F\" target=\"_blank\" rel=\"noopener\">ComputingForGeeks\u003C\u002Fa> 整理的比較表很直接：\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FQwen\" target=\"_blank\" rel=\"noopener\">Qwen 3.5\u003C\u002Fa> 有 256K context，\u003Ca href=\"https:\u002F\u002Fwww.deepseek.com\u002F\" target=\"_blank\" rel=\"noopener\">DeepSeek R1\u003C\u002Fa> 在 MATH-500 拿到 97.3%，\u003Ca href=\"https:\u002F\u002Fwww.zhipuai.cn\u002Fen\u002F\" target=\"_blank\" rel=\"noopener\">GLM-5\u003C\u002Fa> 則在 SWE-bench Verified 拿到 77.8%。講白了，這些數字已經不是「還行」而已，是能進產品討論桌的程度。\u003C\u002Fp>\u003Cp>更現實的是，現在選模型不只看分數。你還得看授權、硬體成本、推理速度，還有能不能合法上線。說真的，這才是開發者每天會撞到的牆。\u003C\u002Fp>\u003Ch2>2026 的開源模型戰場很擠\u003C\u002Fh2>\u003Cp>這份表把主流開源模型幾乎都放進來了。像是 Qwen 3、Qwen 3.5、GLM-5、DeepSeek V3.2、DeepSeek R1、\u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fllama\u002F\" target=\"_blank\" rel=\"noopener\">Llama 4\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fai.google.dev\u002Fgemma\" target=\"_blank\" rel=\"noopener\">Gemma 3\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fmistral.ai\u002F\" target=\"_blank\" rel=\"noopener\">Mistral Large 3\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.cohere.com\u002Fcommand\" target=\"_blank\" rel=\"noopener\">Command A\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.tii.ae\u002F\" target=\"_blank\" rel=\"noopener\">Falcon 3\u003C\u002Fa>，還有 \u003Ca href=\"https:\u002F\u002Fdatabricks.com\u002Fblog\u002Fintroducing-dbrx-new-state-art-open-llm\" target=\"_blank\" rel=\"noopener\">DBRX\u003C\u002Fa>。名單很長，但差異其實很明顯。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775131800331-8pqc.png\" alt=\"2026 開源 LLM 誰領先\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\" target=\"_blank\" rel=\"noopener\">Qwen\u003C\u002Fa> 系列最像全能型選手。Qwen 3.5 397B-A17B 這種架構，雖然總參數很大，但每個 token 只啟動 17B active parameters。這代表推理成本比較好控。對要自己架伺服器的人來說，這點很重要。\u003C\u002Fp>\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-R1\" target=\"_blank\" rel=\"noopener\">DeepSeek R1\u003C\u002Fa> 走的是 MoE 路線，總參數 671B，active parameters 37B。它比較偏向推理。\u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fllama\u002F\" target=\"_blank\" rel=\"noopener\">Llama 4\u003C\u002Fa> 則是把 context length 拉很長，Scout 到 10M tokens，Maverick 到 1M tokens。這種設計很適合長文件、長對話、長程任務。\u003C\u002Fp>\u003Cul>\u003Cli>Qwen 3.5：256K context，支援文字與圖片，Apache 2.0\u003C\u002Fli>\u003Cli>GLM-5：205K context，支援文字與圖片，MIT\u003C\u002Fli>\u003Cli>DeepSeek V3.2：128K context，MIT\u003C\u002Fli>\u003Cli>Llama 4 Maverick：1M context，Llama 4 Community license\u003C\u002Fli>\u003Cli>Mistral Small 4：256K context，Apache 2.0\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Benchmarks 比行銷文案更誠實\u003C\u002Fh2>\u003Cp>看 benchmark，通常比看官網文案更有用。這份比較表用了 MMLU、MMLU-Pro、GPQA Diamond、AIME ’24、MATH-500、SWE-bench Verified。這幾個測試涵蓋常識、進階推理、數學、程式碼修 bug，算是很實在。\u003C\u002Fp>\u003Cp>最亮眼的數字有三個。\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FQwen\" target=\"_blank\" rel=\"noopener\">Qwen 3 235B\u003C\u002Fa> 在 GPQA Diamond 拿到 77.2%，AIME ’24 拿到 85.7%。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-R1\" target=\"_blank\" rel=\"noopener\">DeepSeek R1\u003C\u002Fa> 在 MATH-500 拿到 97.3%，幾乎是把這個測試打到快滿分。\u003Ca href=\"https:\u002F\u002Fwww.zhipuai.cn\u002Fen\u002F\" target=\"_blank\" rel=\"noopener\">GLM-5\u003C\u002Fa> 則在 SWE-bench Verified 拿到 77.8%，是表內最強的 coding 成績。\u003C\u002Fp>\u003Cp>這裡可以借用一句真實的話。\u003Cblockquote>“We a\u003Ca href=\"\u002Fnews\u002Fcrewform-agents-act-like-mcp-tools-zh\">re\u003C\u002Fa> seeing o\u003Ca href=\"\u002Fnews\u002Fopencode-mcp-servers-oauth-support-zh\">pen\u003C\u002Fa> models catch up fast in both quality and efficiency.” — Satya Nadel\u003Ca href=\"\u002Fnews\u002Fsolana-ai-agents-onchain-transactions-99-percent-zh\">la\u003C\u002Fa>, Microsoft Build 2024 keynote\u003C\u002Fblockquote>這句話放到 2026 來看，還是很貼切。開源模型現在的問題，不是能不能做事，而是要做哪件事。\u003C\u002Fp>\u003Cp>另一個值得看的點，是 \u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fllama\u002F\" target=\"_blank\" rel=\"noopener\">Llama 4 Maverick\u003C\u002Fa> 在 MMLU 拿到 85.5%，看起來很漂亮。但 MMLU 只是通用能力的一部分。它不等於深推理，也不等於真的會寫程式。只看單一分數，很容易選錯。\u003C\u002Fp>\u003Cul>\u003Cli>Qwen 3 235B：MMLU-Pro 83.6%，GPQA Diamond 77.2%，AIME ’24 85.7%\u003C\u002Fli>\u003Cli>DeepSeek R1：MMLU-Pro 84.0%，GPQA Diamond 71.5%，MATH-500 97.3%\u003C\u002Fli>\u003Cli>GLM-5：SWE-bench Verified 77.8%\u003C\u002Fli>\u003Cli>Llama 4 Maverick：MMLU 85.5%\u003C\u002Fli>\u003Cli>Gemma 3 27B：MMLU 78.6%，MATH-500 50.0%\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>授權條款，才是能不能上線的分水嶺\u003C\u002Fh2>\u003Cp>很多人先看分數，後看授權。這順序常常會害死人。你模型選得再好，只要法務不給過，產品還是不能出貨。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775131797492-6k09.png\" alt=\"2026 開源 LLM 誰領先\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>目前最省事的，還是 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FQwenLM\u002FQwen\" target=\"_blank\" rel=\"noopener\">Qwen\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-R1\" target=\"_blank\" rel=\"noopener\">DeepSeek\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.zhipuai.cn\u002Fen\u002F\" target=\"_blank\" rel=\"noopener\">GLM-5\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fmistral.ai\u002F\" target=\"_blank\" rel=\"noopener\">Mistral\u003C\u002Fa> 這幾條線。Apache 2.0 和 MIT 對新創最友善。你要 fine-tune、self-host、賣產品，流程都比較乾淨。\u003C\u002Fp>\u003Cp>但 \u003Ca href=\"https:\u002F\u002Fai.meta.com\u002Fllama\u002F\" target=\"_blank\" rel=\"noopener\">Meta 的 Llama\u003C\u002Fa> 就沒那麼單純。Llama 4 和 Llama 3.3 雖然可免費使用，但有 7 億月活用戶門檻。超過之後，就得看 Meta 的條款。\u003Ca href=\"https:\u002F\u002Fai.google.dev\u002Fgemma\" target=\"_blank\" rel=\"noopener\">Gemma\u003C\u002Fa> 則是要接受 Google 條款後才能商用。\u003Ca href=\"https:\u002F\u002Fwww.cohere.com\u002Fcommand\" target=\"_blank\" rel=\"noopener\">Command\u003C\u002Fa> 系列是 CC-BY-NC，商業用途卡得很死。\u003Ca href=\"https:\u002F\u002Fwww.tii.ae\u002F\" target=\"_blank\" rel=\"noopener\">Falcon 3\u003C\u002Fa> 還有營收超過 100 萬美元後的 royalty 條款。\u003C\u002Fp>\u003Cp>所以很多團隊最後選的，不是最強模型，而是最好簽的模型。這很現實，也很台灣。大家都想快上線，但合約常常先把人卡住。\u003C\u002Fp>\u003Cul>\u003Cli>Apache 2.0：Qwen 3\u002F3.5、Mistral Large 3、Mistral Small 4、Mixtral 8x7B、Grok-1\u003C\u002Fli>\u003Cli>MIT：DeepSeek V3\u002FR1\u002FV3.2、Phi-4 變體、GLM-5\u003C\u002Fli>\u003Cli>Llama 4 Community：700M MAU 以下免費，超過後看 Meta 條款\u003C\u002Fli>\u003Cli>CC-BY-NC：Command R+、Command A，不能直接商用\u003C\u002Fli>\u003Cli>DBRX：不能拿去訓練其他 LLM\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>自架測試後，排名會變樣\u003C\u002Fh2>\u003Cp>benchmark 很重要，但跑在自己機器上又是另一回事。這份文章的 Ollama 測試環境很務實。Ubuntu 24.04 LTS，4 vCPUs，16 GB RAM，CPU-only inference。這不是高級 GPU 農場，就是一般開發者比較可能碰到的條件。\u003C\u002Fp>\u003Cp>\u003Ca href=\"https:\u002F\u002Follama.com\u002F\" target=\"_blank\" rel=\"noopener\">Ollama\u003C\u002Fa> 跑 \u003Ca href=\"https:\u002F\u002Fai.google.dev\u002Fgemma\" target=\"_blank\" rel=\"noopener\">Gemma 3 4B\u003C\u002Fa> 時，只用了 4.2 GB RAM。這是表內最省記憶體的模型。\u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fmeta-llama\u002FLlama-3.2-3B\" target=\"_blank\" rel=\"noopener\">Llama 3.2 3B\u003C\u002Fa> 雖然最快，88 秒就回應，但吃了 11.4 GB RAM。\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-R1\" target=\"_blank\" rel=\"noopener\">DeepSeek R1 8B\u003C\u002Fa> 和 \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FQwen\" target=\"_blank\" rel=\"noopener\">Qwen 3 8B\u003C\u002Fa> 都跑到 433 秒，因為推理型模型會先產生更多中間 token。\u003C\u002Fp>\u003Cp>這裡的結論很直接。小模型不一定快，聰明模型常常比較慢。你如果要做本機助理、內網工具、或低成本 API，RAM 和 latency 可能比榜單分數更重要。\u003C\u002Fp>\u003Cul>\u003Cli>Gemma 3 4B：4.2 GB RAM，94 秒\u003C\u002Fli>\u003Cli>Llama 3.2 3B：11.4 GB RAM，88 秒\u003C\u002Fli>\u003Cli>Phi-4 Mini 3.8B：8.9 GB RAM，97 秒\u003C\u002Fli>\u003Cli>Mistral 7B：7.4 GB RAM，125 秒\u003C\u002Fli>\u003Cli>Qwen 3 8B、DeepSeek R1 8B：都要 433 秒\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>我會怎麼選\u003C\u002Fh2>\u003Cp>如果是我今年要上產品，我會先看 \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002FQwen\" target=\"_blank\" rel=\"noopener\">Qwen 3.5\u003C\u002Fa>。它的泛用性高，context 也夠長。要做推理任務，我會看 \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002FDeepSeek-R1\" target=\"_blank\" rel=\"noopener\">DeepSeek R1\u003C\u002Fa>。要做 coding，我會先試 \u003Ca href=\"https:\u002F\u002Fwww.zhipuai.cn\u002Fen\u002F\" target=\"_blank\" rel=\"noopener\">GLM-5\u003C\u002Fa>，因為 SWE-bench Verified 的數字很漂亮。\u003C\u002Fp>\u003Cp>但真正的選型邏輯，不是「誰最強」。而是「誰最適合你的工作」。如果你是法規很重的企業，Apache 2.0 或 MIT 幾乎是首選。如果你要處理超長文件，Llama 4、Qwen 3.5、Mistral Large 3 都值得測。如果你在意程式碼修補，GLM-5 要先進你的測試清單。\u003C\u002Fp>\u003Cp>我覺得 2026 的重點很簡單。開源模型已經能打進實戰，但真正拉開差距的，是你的資料、你的提示詞、你的部署方式，還有你能不能快速換模型。這件事很多團隊還沒準備好。\u003C\u002Fp>\u003Ch2>接下來該看什麼\u003C\u002Fh2>\u003Cp>如果你現在要做選型，別只看一張排行榜。先拿自己的資料跑 20 到 50 個真實任務。再比 latency、RAM、成本和授權。這樣比看新聞稿準多了。\u003C\u002Fp>\u003Cp>我會猜，接下來 6 到 12 個月，開源 LLM 的競爭焦點會更偏向「同級效能下的成本」和「授權條款」。誰能把推理成本壓低，誰就更容易進企業環境。你如果是開發者，現在就該把模型切換流程做成可插拔，不然之後會很痛。\u003C\u002Fp>\u003Cp>說白了，2026 的問題不是開源模型能不能用。問題是，你的產品能不能跟著換。這才是現在最值得先處理的事。\u003C\u002Fp>","Qwen 3.5、GLM-5、DeepSeek R1、Llama 4 讓開源 LLM 進入實戰。這篇整理 2026 年主流模型的 benchmark、上下文長度、授權條款與自架表現。","computingforgeeks.com","https:\u002F\u002Fcomputingforgeeks.com\u002Fopen-source-llm-comparison\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1775131800331-8pqc.png","model-release","zh","424af64f-8d0b-4cd5-b58b-f37ee073bfa1",[17,18,19,20,21,22,23,24,25,26],"開源 LLM","Qwen 3.5","DeepSeek R1","GLM-5","Llama 4","模型比較","授權條款","SWE-bench Verified","MATH-500","自架部署",23,"2026-04-02T12:09:39.445524+00:00","2026-04-02T12:09:39.227+00:00",{"tags":31,"relatedLang":40,"relatedPosts":44},[32,34,37,38],{"name":21,"slug":33},"llama-4",{"name":35,"slug":36},"DeepSeek-R1","deepseek-r1",{"name":22,"slug":22},{"name":24,"slug":39},"swe-bench-verified",{"id":15,"slug":41,"title":42,"language":43},"open-source-llm-comparison-2026-en","Open Source LLMs in 2026: Who Leads?","en",[45,51,57,63,69,75],{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":13},"cfb68e08-fe4e-49f6-b449-e566faf56311","kimi-2-7-price-coding-benchmark-zh","Kimi 2.7 讓價格成為真正的寫碼基準","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782746270225-tcs9.png","2026-06-29T15:17:24.321277+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":13},"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","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782739078329-qvne.png","2026-06-29T13:17:26.530857+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":13},"edf8e66b-c717-4cc1-b15a-96839bb7bbcf","llama-legends-380-season-3-heroes-raids-zh","Llama Legends 3.8.0 推出 Season 3 英雄與突襲","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782711179415-qurv.png","2026-06-29T05:32:32.733919+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":13},"88d353ca-468b-4774-922d-ef0cbc2edd68","omlx-045-dev1-glm52-minimax-m3-speedups-zh","oMLX 0.4.5.dev1 讓長上下文更快","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782709372375-25nm.png","2026-06-29T05:02:28.341041+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":13},"e6ae84b6-4e55-4ab2-a1cf-4a08e23cbc77","grok-45-private-beta-tesla-spacex-zh","Grok 4.5 先進 Tesla 和 SpaceX 內測","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782687769532-te5b.png","2026-06-28T23:02:22.915901+00:00",{"id":76,"slug":77,"title":78,"cover_image":79,"image_url":79,"created_at":80,"category":13},"186b266a-5b45-4bd4-85a4-5fa62fcc50dc","google-openrl-llm-fine-tuning-kubernetes-zh","Google OpenRL 把 RL 細調搬上 Kubernetes","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1782572576166-gzxw.png","2026-06-27T15:02:27.036919+00:00",[82,87,92,97,102,107,112,117,122,127],{"id":83,"slug":84,"title":85,"created_at":86},"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":88,"slug":89,"title":90,"created_at":91},"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":93,"slug":94,"title":95,"created_at":96},"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":98,"slug":99,"title":100,"created_at":101},"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":103,"slug":104,"title":105,"created_at":106},"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":108,"slug":109,"title":110,"created_at":111},"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":113,"slug":114,"title":115,"created_at":116},"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":118,"slug":119,"title":120,"created_at":121},"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":123,"slug":124,"title":125,"created_at":126},"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":128,"slug":129,"title":130,"created_at":131},"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"]