Alibaba Cloud180B (6B active)Q4_K_M 下约 109 GB 显存

Qwen3.8-Flash-Next — 显存、速度与本地部署

作者： Jakub Rusinowski · 最后更新：

An experimental preview of the architecture intended to underpin Qwen4, not a conventional production release. 180B of parameters must be resident — a 125B MoE backbone, a 51B n-gram lookup table and a 4B multi-token-prediction head, about 109 GB at Q4_K_M — while only 6B activate per token. 48 layers alternating three Gated DeltaNet blocks (compressing history into a fixed recurrent state) with one Qwen Sparse Attention block (precise long-range retrieval). Natively multimodal, 262,144-token context extensible to 1M via YaRN. Qwen Community License 1.0, NOT the Apache 2.0 the 27B carries.

Qwen3.8-Flash-Next 在 Q4_K_M 下约需 109 GB 显存——量化权重加框架开销，不含 KV 缓存。在 Apple Silicon 上，这部分来自统一内存。

逻辑89
创意84
编程90

按量化级别的显存与速度

计算基准：NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存，因此数值高于上方的主数字。

量化显存速度（估算）适配
Q2_K
2.63 bpw
60.2 GB—放不下
Q3_K_M
3.41 bpw
77.7 GB—放不下
Q4_K_M
4.83 bpw
109.7 GB—放不下
Q5_K_M
5.67 bpw
128.6 GB—放不下
Q6_K
6.56 bpw
148.6 GB—放不下
Q8_0
8.50 bpw
192.3 GB—放不下
F16
16.00 bpw
361 GB—放不下

黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型，详见 方法说明页. Qwen3.8-Flash-Next 显存计算器 →

运行 Qwen3.8-Flash-Next

目录中能运行 Qwen3.8-Flash-Next 的最便宜 GPU 是 NVIDIA DGX Spark (128 GB).

购买此硬件 NVIDIA DGX Spark (128GB) — 128 GB VRAM · 150 W board power立即云端部署 RunPod

或在 Vast.ai 比较

作为亚马逊联盟成员，我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金，您无需额外付费。

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买，LLM Configurator 可能会获得佣金，而你无需支付任何额外费用。作为亚马逊联盟成员（Amazon Associate），LLM Configurator 会从符合条件的购买中获得收益。
NVIDIA DGX Spark (128GB)
128 GB VRAM · 150 W board power
2026年价格波动较大——请以当前商品页价格为准。

如何运行 Qwen3.8-Flash-Next

安装 Ollama，然后运行：

ollama run qwen3-8
Hugging Face 上的权重： Qwen/Qwen3.8-Flash-Next ↗

规格

Verified — Checked against the primary source — the model card or the vendor spec page — and corroborated by a second independent source.

参数量
180 Billion (6B active)
上下文窗口
262,144
架构
Gated DeltaNet + Qwen Sparse Attention MoE
提供商
Alibaba Cloud
许可证
Qwen Community License 1.0
Practical deployment
Workstation (large unified memory or pro GPU)
规格量化
Q4_K_M
系统内存
192 GB
记录更新于
2026-09-11
许可证Qwen Community License 1.0允许商业使用

Weights are downloadable and commercial use is permitted, subject to the licence’s acceptable-use terms.

Above 100M monthly active users or $20M monthly revenue, an attribution duty applies; model-as-a-service resale needs a separate agreement.

质量与使用场景

评分由模型作者或独立评测方发布——衡量质量而非吞吐量，并非我们实测。

最适合researchagentic codinglong contextmultimodal

我的 GPU 能运行 Qwen3.8-Flash-Next 吗？

Qwen3.8 的其他尺寸

Qwen3.8-Flash-Next — 常见问题

How much VRAM does Qwen3.8-Flash-Next need?

About 109 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does Qwen3.8-Flash-Next run on an RTX 4090 (24 GB)?

No. Qwen3.8-Flash-Next needs about 109 GB at Q4_K_M, more than a single RTX 4090's 24 GB. It needs a larger card, several GPUs, or Apple Silicon with enough unified memory — or it runs with part of the weights offloaded to system RAM, which is much slower.

How do I run Qwen3.8-Flash-Next locally?

Install Ollama and run `ollama run qwen3-8`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does Qwen3.8 come in?

Qwen3.8 27B (18 GB), Qwen3.8-Max (1450 GB), Qwen3.8-Flash-Next (109 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.