Skip to main content
← Back to news

Qwen3.8-Flash-Next: the 125B MoE with only 6B active parameters, built for agentic work

Model releases
Source:Hugging Face (Qwen)
Alibaba releases Qwen3.8-Flash-Next: 125B parameters with only 6B activated, plus a 51B N-gram embedding and 4B MTP. The hybrid architecture alternates 48 layers between Gated DeltaNet and Qwen Sparse Attention (512 experts, 10 routed + 1 shared), with a native 262K-Token context extensible to 1M. It scores 58.7 on DeepSWE 1.1 — the model is built for the agentic workloads that dominate real usage.
Read original article

Related articles

Qwen3.8-27B vs Gemma 4 26B A4B: the new frontier at 24GB

A 27.8B dense or a MoE with 3.8B active, a 262K context against 256K and almost the same weight at Q4_K_M: two opposite ways of putting big-model quality on a 24GB consumer card. Which is faster, which goes deeper, and when to pick one over the other.

Ling 3.1 Flash: 560B open-weight with MIT license and 1M context — what changes if you have 24GB of VRAM

Ling 3.1 Flash ships 560B open-weight parameters under an MIT license with 1M context. It won't fit in 24GB of VRAM, but the quantizations, 25B active MoE and the real numbers change the calculation.

Llama 3 and Llama 4: Meta's open family — which model runs on your hardware

From the 3.1 8B to the 4 Scout MoE with a 10M context: the Llama family in a table, what runs on 16, 48 and 128 GB of RAM, the Community license translated for the Italian business and the choice by budget.

Get the next local-AI briefing

Once a week, the top selection from iaLocale: articles, news and resources on local AI, and more.

After signing up you receive a confirmation email (double opt-in). Without confirmation we send nothing. No spam, we promise.

Related resources

Tools and resources from the iaLocale archive relevant to this content.

🦉

Qwen3.8-Flash-Next

Models

Open-weight MoE of the Qwen3.8 family built for agentic work: 125B parameters with only 6B activated, plus a 51B n-gram embedding and 4B MTP. Hybrid attention over 48 layers (12 micro-blocks: 3× Gated DeltaNet → MoE, 1× Qwen Sparse Attention → MoE), 512 experts with 10 routed + 1 shared. Native 262K context, extensible to 1M with RoPE scaling (YaRN).

🧠

Qwen3 30B-A3B (MoE)

Models

Mixture-of-Experts model: 30B total params but only 3.3B active per token. Small-model speed with mid-size quality; the best speed/quality ratio in its class.

🌏

inclusionAI Ling 3.1 Flash (MoE)

Models

Modello MoE open-weight di inclusionAI: 560B parametri totali, 25B attivi per token, architettura ibrida (attenzione lineare KDA + Gated MLA), finestra di contesto fino a 1M di token, licenza MIT. Reasoning ibrido con modalità di pensiero on/off. A 4 bit entra in 512GB di memoria unificata.

⚡

DeepSeek V4.1 Flash

Models

DeepSeek's MoE model with 552B total parameters and 8-16B active per token. 1M-token context, MIT license: free commercial use. Built for online serving, too large for local use on a single PC.

Browse all resources