Simon Willison experiments with Qwen3.8-Flash-Next, a new open-weight multimodal MoE model that serves as an early preview of the architecture used in Qwen4. The model has 125B total parameters but only 6B active, offering strong performance while running on a DGX Spark using Unsloth quantized versions.
Background
Qwen is Alibaba's series of open-weight large language models. The MoE (Mixture of Experts) architecture with sparse activation has become a key approach for balancing performance and inference cost in modern LLMs.
- Source
- Simon Willison
- Published
- Aug 27, 2026 at 07:52 AM
- Score
- 6.0 / 10