# Qwen3.8-27B 发布：27B 参数原生多模态模型，综合表现超越 Qwen3.7-Plus

- 来源：Hacker News 热门（buzzing.cc 中文翻译）
- 作者：mfiguiere
- 发布时间：2026-08-14 23:39
- AIHOT 分数：67
- AIHOT 链接：https://aihot.virxact.com/items/cmst4kkgf04gbro06olkgg0da
- 原文链接：https://twitter.com/alibaba_qwen/status/2088280182356611304

## AI 摘要

通义千问发布 Qwen3.8-27B，一款仅 27B 参数的原生多模态密集模型，综合表现超越 Qwen3.7-Plus，在真实编码与办公工作流中尤为突出。该模型原生支持 262K 上下文，可通过 YaRN 轻松扩展至 1M token，采用 Apache 2.0 许可证开源。此外，Qwen3.8-2.4T-A95B（Max 级）的开放权重也已于近期发布。

## 正文

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://t.co/4kaAcqYEVj

ModelScope： https://t.co/eRIMZCGkhC