MODEL LINEAGE
中文
Model index
Alibaba Cloud·Qwen

Qwen3.8-Flash-Next

A Qwen4 architecture preview: 125B main model plus 51B N-gram embeddings, 6B active per token, extendable to 1M context.

CurrentReasoningCodingAgentMultimodalOpen weights
MODEL NUMBER116
First public date
2026-08-26
Organization
Alibaba Cloud
Family
Qwen
01

Specifications

Only published specifications are shown.

Total parameters
176B↗
Active parameters
6B↗
Architecture
Hybrid↗
Parameter scope
The roughly 125B main model, 51B N-gram embedding, and MTP components are separate scopes; about 6B is active per token.↗
Context
262,144 tokens
Input modalities
Text · Image
Open weights
Yes
License
Apache-2.0
Status
Current
02

Overview

Release dates

— Qwen3.8-Flash-Next released open weights and became available as managed Qwen3.8-Flash. ↗

Why it mattered

It introduced QSA, Gated Residual, and N-gram memory to reduce compute while advancing long-context agents.

03

Architecture & capabilities

Attention

Gated DeltaNetQwen Sparse Attention

Model capabilities

CodingAgentLong context

API features

Adjustable reasoning
04

Release history

Open weights

Qwen3.8-Flash-Next released open weights and became available as managed Qwen3.8-Flash.

View source
05

Related models

Select a card to view the related model.

SiblingQwen3.8-Max APIView details →Derived modelQwen3.8-Omni-FlashThe official release says it is built on the Qwen3.8-Flash-Next architectureView details →
06

Sources

01
Qwen3.8-Flash-NextQwen Team · Published 2026-08-26
Official · Verified02
Qwen3.8-Flash-Next model cardQwen Team · Checked 2026-09-07
Official · Verified
Suggest a correction

Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.