🤯我靠！NVIDIA 直接把 Qwen3.8-Flash-Next 砍到原来的 37% 体积！这是 Blackwell 时代真正能跑起来的实用级效率！
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！NVIDIA 直接把 Qwen3.8-Flash-Next 砍到原来的 37% 体积！这是 Blackwell 时代真正能跑起来的实用级效率！ 125B MoE + 6B 激活 + 混合注意力（GDN+QSA）+ 51B N-gram Embedding，原版 BF16 约 360GB，NVFP4 版约 2.7x 更小，精度几乎没掉。 关键亮点： 🔹比 Claude Opus 4.6 Max 在 SWE-bench Pro、CoWorkBench、JobBench 等真实任务上更好 🔹原生 262K 上下文，可扩展到 1M 🔹支持文本/图像/视频多模态 🔹路由专家 NVFP4 + 关键层 BF16/FP8 混合精度，精度损失极低 关键使用步骤： 1️⃣确认硬件：推荐 NVIDIA Blackwell（B200/B300），Linux 环境 2️⃣下载模型：huggingface-cli download nvidia/Qwen3.8-Flash-Next-NVFP4 3️⃣安装支持的 vLLM（需较新 commit）+ Transformers ≥ 5.16.0 4️⃣启动命令示例： vllm serve nvidia/Qwen3.8-Flash-Next-NVFP4 \ --tensor-parallel-size 8 \ --
Posted by SuSu_酥酥👅 (24.5k followers) 3 days ago · 185 likes · 20k views · view the original post on X. Kept by the Dev Radar as AI dev tools. Tools mentioned: qwen3.8-flash-next, qwen3.8-flash-next-nvfp4, opengcua, arXiv.org.
More dev work like this
- 🎃Dario给中国人民带来的极致恐怖从Claude无情封号开始 — @NFT_Chen
- Mia's Qwen3.8-Flash-Next recipe on a DGX Spark: one command, first try, 2 min to the… — @niklaslenz_ai
- 给AI喂一整仓，最烦的就是它每回都从头翻文件，这毛病你多半也遇到过。 — @bkdgiffug
- Coding agents aren't slow because they're dumb. They're slow because every edit waits on… — @RunAnywhereAI
- ICYMI: Foursquare co-founder @dens sat down with our Head of Product, AI, Jeff Smith at… — @automattic
- 🤯惊了！9B参数大模型，只需16G内存即可本地跑通，能力反超Qwen3.5-27B！ — @NFT_Chen
- Introducing JetSpec++: we find causal parallel drafting paired with hidden states… — @haoailab
- Build the agent, test the agent, grade the results, change the agent, run the test cases… — @rseroter
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 2026-10-30 20:27 UTC. Full method.