🤔我靠！微软悄咪咪开源4B AI代理Mage！66%逼近人类72.4%，碾压6倍大模型，本地可跑隐私拉满！
This is a dev post classified by AGK Intelligence as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
🤔我靠！微软悄咪咪开源4B AI代理Mage！66%逼近人类72.4%，碾压6倍大模型，本地可跑隐私拉满！ 微软刚放Mage-VL-Cowork：4B开源混合计算机使用代理，OSWorld-Verified拿下66%（人类72.4%），直接打赢几乎6倍大的Qwen3.6-27B（63.9%）。 它自己决定何时像人一样点GUI、何时直接敲终端命令，一步完成复杂桌面任务。电子表格39秒搞定，大模型点10分钟还没完。9.5GB体量，Mac mini轻松跑，截图不出本地，Apache 2.0可商用。 核心亮点 🔹极致高效：4B规模SOTA级，在线RL后模型调用少13%、输出token少25.4%，真正推动端侧智能。 🔹混合决策：单一策略同时掌握点击/拖拽和bash终端，无需外部调度。 🔹开源友好：基于Mage-ViT + Qwen3-4B，32K上下文，完整训练配方和评估全开源。 🔹实测落地：真实任务一次过，本地隐私保护到位。 使用步骤 1️⃣Hugging Face下载：microsoft/Mage-VL-Cowork（约9.5GB）。 2️⃣安装依赖：pip install "transformers>=4.57.3" accelerate pillow torch torchvision qwen-vl-utils（推荐SGLang/Docker）。 3️⃣加载加t
Posted by SuSu_酥酥👅 (25.1k followers) 49 min ago · 15 likes · 1.1k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: mage, mage-vl-cowork, BenchPress.
More dev work like this
- Gemini 4 Carbon scores 53 FCI and 54 pass rate on FrontierCS 2. It’s an impressive jump… — @wenhaocha1
- Clover’s speed caught my attention. Its task results deserve attention too. — @TeksEdge
- This kinda went viral. — @MengTo
- 🤨匪夷所思！Gemini 4 Carbon 在 FrontierCode 2.0 只拿 40%，居然还落后 Sonnet 5.5 的 41%！ — @NFT_Chen
- 花钱找人做SEO诊断，和用开源工具jev-seo一分钟爬完整站，我选后者。 — @bkdgiffug
- B e f o r e A n y o n e K n e w 🎞️ — @juliafedorin
- 𝗛𝗲𝗿𝗺𝗲𝘀 Agent can read your Google Analytics and answer in plain English. 67 tools,… — @BkashJosi
- An RTX 2080 Ti from 2018. Modded to 22GB. — @TeksEdge
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.2k posts from 5k X accounts over the last 21 days, 8.8k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-10-31 06:37 UTC. Full method.