Summary & Key Takeaways ​

• Qwen3.8-Flash-Next is a new open-weights multimodal Mixture-of-Experts (MoE) model.
• It serves as an early preview of the architecture planned for Qwen4.
• The model is large (125B tokens) but has only 6B active parameters, boosting performance.
• Simon Willison shares his initial experiences trying out quantized versions of the model.
• This release is significant for the open-source AI community and model architecture research.

Our Commentary ​

A new open-weights multimodal MoE model from Qwen, and an early peek at Qwen4's architecture? That's big news for the AI community. The 125B total, 6B active parameter count is intriguing for performance. Simon Willison's early explorations are always a good sign of something interesting brewing.