阿里巴巴发布Qwen3.8-27B，一款27B参数的开源多模态稠密模型，采用Apache 2.0许可，支持Transformers、vLLM、SGLang及本地量化部署。
Alibaba dropped the weights for Qwen3.8-27B as a 27B open-weight multimodal model built for local deployment.
- Apache 2.0 weights and support for Transformers, vLLM, SGLang, and local quantizations, Qwen3.8-27B puts unusually capable multimodal agent work within single-machine deployment range.
- It has 262k tokens of native context, extendable to 1M with YaRN, while reasoning can be disabled or adjusted per request.
- AMD says Qwen3.8-27B reached up to 51.8 tokens/sec in its initial testing on a single Radeon AI PRO R9700; roughly 24GB VRAM
- For coding, Qwen3.8-27B surprisingly close to, and sometimes above, Claude Opus 4.6 Max:
SWE-bench Pro 61.7 vs 53.4, CoWorkBench 70.7 vs 68.2, and OSWorld 84.3 vs 72.7.
So this 27B local model is legitimately frontier-class on several coding/agent benchmarks
We promised open weights for Qwen3.8. Now, time to meet them! 🎉 ⚡ Qwen3.8-27B: - A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows. - 262K native context, easily extendable to 1M tokens via YaRN. - Built for builders. Highly efficient, high-quality, and licensed under Apache 2.0. 🚀 The open weights for Qwen3.8-2.4T-A95B (Max-level) have also been released recently. Whether you're shipping lightweight applications with Qwen3.8-27B locally or building agents with Qwen3.8-2.4T-A95B, they're yours now! Download, deploy, and build something we haven't imagined yet. 👀👇 - Hugging Face： https://huggingface.co/collections/Qwen/qwen38 - ModelScope： https://www.modelscope.cn/collections/Qwen/Qwen38在 X 查看被引用的帖子
来源：Rohan Paul · x.com