Qwen3.8-Flash-Next

A 180B mixture-of-experts model that uses only 6B per token, and a preview of the architecture planned for Qwen4.

In short. Qwen3.8-Flash-Next is a fast, mid-sized AI model from Alibaba. Although it stores 180 billion parameters, it only works with 6 billion at a time, so it answers quickly. It also shows the new design that Alibaba plans to use in its next generation, Qwen4.

Summary

Qwen3.8-Flash-Next was released on 26 August 2026. Alibaba presents it as an early public preview of the architecture that will underpin Qwen4, in the same way that Qwen3-Next previewed the design later used across Qwen3.5 to Qwen3.8.

It is a multimodal mixture-of-experts model: a 125-billion-parameter main model with 6 billion parameters active per token (10 of 512 experts plus one shared expert), plus 51 billion parameters of n-gram embeddings and a 4-billion-parameter multi-token-prediction layer, for 180 billion in total. It accepts text and images and writes text. The context window is 262,144 tokens natively and extensible to 1 million.

Four ideas are new. A sparse-attention layer called Qwen Sparse Attention works on blocks of tokens and is paired with Gated DeltaNet. Gated Residual widens the residual stream. N-gram embeddings add capacity through lookup tables that are cheap to compute and easy to offload. A new training recipe uses the Muon optimizer for some weights. Alibaba says training cost about one ninth of Qwen3.7-Plus while performing better on coding and office tasks; this is a vendor claim.

Weights come as BF16 (about 360 GB) and FP8 (about 186 GB). The license is the Qwen Community License 1.0.

Good for

  • Fast, cheap serving of a capable multimodal assistant on a single multi-GPU server
  • Agent and coding workloads where only a few billion active parameters per token keeps latency low
  • Researchers who want to study the architecture expected in Qwen4
  • Long-context work up to 1 million tokens

Watch out for

  • All 180 billion parameters must be in memory even though only 6 billion are used per token
  • The Qwen Community License 1.0 is not Apache 2.0; read its conditions before commercial use
  • It is a preview of a new architecture, and tooling support may lag behind older models
  • Alibaba's cost and quality comparisons are self-reported

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Estimates for the Flash-Next variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.

Variants Parameters BF16 FP8 / INT8 4-bit (Q4) Fits on
Flash-Next BF16 180B 6B active 414 GB 207 GB 124 GB Big-memory desktop
Flash-Next FP8 FP8 180B 6B active 414 GB 207 GB 124 GB Big-memory desktop

Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.

VRAM calculator

The license in plain words

An MIT-style license with a stricter hosting rule than the Qwen3.8-Max License. You can build products with the model freely, but anyone running a model-as-a-service or AI work-assistant business needs a separate license, whatever their revenue.

  • Attribution or model name must be shown
  • Separate deal needed to sell hosted access above a threshold

More about Qwen Community 1.0

See also

Frequently asked questions

What is Qwen3.8-Flash-Next?
Qwen3.8-Flash-Next is a fast, mid-sized AI model from Alibaba. Although it stores 180 billion parameters, it only works with 6 billion at a time, so it answers quickly. It also shows the new design that Alibaba plans to use in its next generation, Qwen4.
Can I use Qwen3.8-Flash-Next commercially?
Yes, with conditions. Qwen3.8-Flash-Next is released under the Qwen Community 1.0, which allows commercial use but adds extra rules: Attribution or model name must be shown; Separate deal needed to sell hosted access above a threshold. Read the license text before you ship a product.
Is Qwen3.8-Flash-Next open source?
Qwen3.8-Flash-Next is open-weight, not open source in the strict sense. You can download the model files, but the Qwen Community 1.0 adds conditions that the Open Source Initiative definition does not allow.
Can I run Qwen3.8-Flash-Next on my own computer?
Only on a machine with a lot of memory. The smallest variant, Flash-Next, needs about 124 GB at 4-bit precision. Desktops with 128 to 512 GB of unified memory can hold it, at modest speed.
When was Qwen3.8-Flash-Next released?
Qwen3.8-Flash-Next was released by Alibaba Qwen on August 26, 2026.
How long a context window does Qwen3.8-Flash-Next have?
Qwen3.8-Flash-Next supports a context window of up to 256K tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 8, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Qwen3.8-27B Alibaba Qwen 27.8B 256K Apache 2.0 (Permissive) 19 GB Gaming GPU
GLM-5.3-Flash Z.ai (Zhipu AI) 320B 18B active 1M MIT (Permissive) 221 GB Big-memory desktop
MiniMax M3.1 Flash MiniMax 428B 23B active 1M MiniMax Community (Conditional) 277 GB Big-memory desktop
Qwen-Image-2.1 Alibaba Qwen 7.1B – Qwen Research License (Non-commercial) 20 GB Gaming GPU