---
id: 20260827-T0-01
title: "Qwen3.8-Flash-Next发布：125B参数MoE，仅6B激活"
title_en: "Qwen3.8-Flash-Next: 125B MoE Model with only 6B Active Parameters"
url: https://ai.daily.yangsir.net/daily/20260827-T0-01
issue_date: 2026-08-27
publish_date: 2026-08-26T23:52:58.000Z
category: release
source_name: "Simon Willison"
source_url: https://simonwillison.net/2026/Aug/26/qwen38-flash-next/
---

# Qwen3.8-Flash-Next发布：125B参数MoE，仅6B激活

Qwen开源新模型Qwen3.8-Flash-Next，这是一个多模态MoE模型，也是Qwen4架构的早期预览版。模型总参数125B，但每次推理仅激活6B参数，带来显著性能提升。开发者Simon Willison已在DGX上进行了测试。该模型适合需要高吞吐、低延迟的推理场景。

## English Version

**Qwen3.8-Flash-Next: 125B MoE Model with only 6B Active Parameters**

Qwen released Qwen3.8-Flash-Next, a multimodal MoE model and early preview of the Qwen4 architecture. With 125B total parameters but only 6B active per inference, it offers significant performance gains. Developer Simon Willison has tested it on DGX, making it suitable for high-throughput, low-latency inference.

---

**来源**：[Simon Willison](https://simonwillison.net/2026/Aug/26/qwen38-flash-next/)

**详情页**：https://ai.daily.yangsir.net/daily/20260827-T0-01

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*