「本地 Opus 4.6」呢個稱號,反映佢喺部分編程、辦公代理同電腦操作測試表現突出;但現有證據未足以證明佢喺所有推理及代理任務全面追平雲端前沿模型。
What is Alibaba Qwen3.8-27B, the open-source Apache 2.0 model that topped Hugging Face’s global trending chart within two days, exceeded oneQwen3.8-27B’s headline is not universal frontier-model parity, but the ability to bring strong coding and computer-use capabilities to local hardware.
AI 提示
Create a landscape editorial hero image for this Studio Global article: What is Alibaba Qwen3.8-27B, the open-source Apache 2.0 model that topped Hugging Face’s global trending chart within two days, exceeded one. Article summary: Alibaba Qwen3.8-27B is a 27-billion-parameter, dense, open-weight multimodal model aimed at local agentic coding, tool use, and computer-use workloads. Its “local Opus 4.6” nickname is directionally useful for selected b. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
openai.com
一部可以喺消費級硬件運行嘅 27B 模型
Alibaba 旗下 Qwen3.8-27B 係一個 270 億參數嘅 dense(密集式)開放權重多模態模型,主打編程、推理、工具調用,以及電腦操作代理工作流。佢最吸引開發者嘅地方,唔單止係基準測試分數,而係經過量化之後,可以喺遠低於一般前沿 AI 所需規模嘅硬件上運行。14
模型推出後迅速成為本地 AI 圈熱話。Qwen3.8 主要模型庫喺幾日內突破一百萬次下載,Unsloth 嘅量化版本其後亦報稱錄得 270 萬次下載。不過,呢啲係會隨時間變化嘅平台數據,下載量本身唔等於模型質素證明。485456
最值得留意嘅唔係一個「總冠軍」標籤,而係模型嘅專長分布。當任務要求模型跟足指示、寫或者修復程式、操作電腦,或者完成一套有結構嘅工作流程時,Qwen3.8-27B 表現特別突出。佢喺 SWE-bench Pro 得分 61.7,高過報告中 Opus 4.6 Max 嘅 53.4;CoWorkBench 亦以 70.7 輕微高於 68.2。34
「本地 Opus 4.6」呢個稱號,反映佢喺部分編程、辦公代理同電腦操作測試表現突出;但現有證據未足以證明佢喺所有推理及代理任務全面追平雲端前沿模型。
What is Alibaba Qwen3.8-27B, the open-source Apache 2.0 model that topped Hugging Face’s global trending chart within two days, exceeded oneQwen3.8-27B’s headline is not universal frontier-model parity, but the ability to bring strong coding and computer-use capabilities to local hardware.
AI 提示
Create a landscape editorial hero image for this Studio Global article: What is Alibaba Qwen3.8-27B, the open-source Apache 2.0 model that topped Hugging Face’s global trending chart within two days, exceeded one. Article summary: Alibaba Qwen3.8-27B is a 27-billion-parameter, dense, open-weight multimodal model aimed at local agentic coding, tool use, and computer-use workloads. Its “local Opus 4.6” nickname is directionally useful for selected b. Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
openai.com
一部可以喺消費級硬件運行嘅 27B 模型
Alibaba 旗下 Qwen3.8-27B 係一個 270 億參數嘅 dense(密集式)開放權重多模態模型,主打編程、推理、工具調用,以及電腦操作代理工作流。佢最吸引開發者嘅地方,唔單止係基準測試分數,而係經過量化之後,可以喺遠低於一般前沿 AI 所需規模嘅硬件上運行。14
模型推出後迅速成為本地 AI 圈熱話。Qwen3.8 主要模型庫喺幾日內突破一百萬次下載,Unsloth 嘅量化版本其後亦報稱錄得 270 萬次下載。不過,呢啲係會隨時間變化嘅平台數據,下載量本身唔等於模型質素證明。485456
最值得留意嘅唔係一個「總冠軍」標籤,而係模型嘅專長分布。當任務要求模型跟足指示、寫或者修復程式、操作電腦,或者完成一套有結構嘅工作流程時,Qwen3.8-27B 表現特別突出。佢喺 SWE-bench Pro 得分 61.7,高過報告中 Opus 4.6 Max 嘅 53.4;CoWorkBench 亦以 70.7 輕微高於 68.2。34