Qwen 3.8-Flash-Next Release

Qwen 3.8-Flash-Next introduces next-generation Qwen4 architecture

Alibaba is releasing Qwen 3.8-Flash-Next, a multimodal Mixture-of-Experts (MoE) model. This release serves as an early preview of the architectural advancements that will be integrated into the upcoming Qwen4 model family, allowing the community to prepare for the full family release.

Model Specifications and Availability

Qwen 3.8-Flash-Next is scheduled for release on August 26, 2026, at 15:00 UTC. The release includes two primary versions of the model:

• Qwen/Qwen3.8-Flash-Next: The standard open model.
• Qwen/Qwen3.8-Flash-Next-FP8: A quantized version for improved efficiency.

While the official ModelScope page does not explicitly list the parameter count in the main text, community discussion indicates a configuration of 125B total parameters with 6B active parameters (125B a6B).

Local Deployment and Hardware Requirements

The MoE architecture of Qwen 3.8-Flash-Next makes it a target for users with high-memory consumer hardware. Because MoE models only activate a fraction of their total parameters during inference, they can potentially achieve usable speeds on local machines if sufficient RAM/VRAM is available to hold the model weights.

Community members have highlighted several hardware configurations for potential deployment:

• High-RAM Systems: Users with 128GB of RAM (such as those using Strix Halo or Mac Studio/M-series Max chips) anticipate that 4-bit quantizations with large context windows will fit and perform efficiently.
• Hybrid Memory Layouts: Some users report success running MoE models by placing dense reasoning layers on the GPU and experts on the CPU, achieving speeds of up to 40 tokens per second on previous 80B-a3B models.
• Inference Optimizations: There is significant interest in the intersection of this model with inference engines like FreeToken, which aim to optimize the distribution of work between CPU/RAM and GPU/VRAM for MoE architectures.

Community Perspectives and Market Positioning

Developers and enthusiasts are monitoring how Qwen 3.8-Flash-Next compares to other high-efficiency models, specifically DeepSeek v4 Flash. The early release of architectural improvements is seen as a strategic move to provide a community preview before the full Qwen4 rollout.

Some developers have noted challenges with the stability and capacity of Qwen models when accessed via third-party providers like OpenRouter, suggesting that while the model architecture is impressive, the infrastructure for reliable API access remains a pain point for some users.

"I enjoy the Qwen models a lot, but building things on top of them with OpenRouter has been painful... many Qwen models have almost no capacity or is so flaky you literally have to just litter your code with a blacklist/whitelist of providers."

Overall, the release is viewed as a signal of Alibaba's continued aggressive pace in the open-weights model space, potentially challenging the dominance of proprietary models for coding and technical tasks.