Qwen releases Qwen3.8-Flash-Next as open-weights multimodal MoE previewing Qwen4 architecture
The 125B-token model uses a 6B-active sparse MoE to deliver a performance boost while previewing architecture slated for Qwen4.
1 source · cross-referenced
- Qwen released Qwen3.8-Flash-Next as an open-weights multimodal mixture-of-experts model.
- The model has 125B total tokens with only 6B active parameters, enabling a significant performance boost.
- It serves as an early preview of the architecture planned for Qwen4.
- Simon Willison reports hands-on testing on a DGX Spark using Unsloth quantized versions of the model.
Qwen released Qwen3.8-Flash-Next as an open-weights model described as a multimodal mixture-of-experts (MoE) architecture and positioned as an early preview of the design that will underpin Qwen4.
The model’s scale is reported at 125B total tokens, with only 6B active parameters, a configuration intended to deliver a significant performance boost through sparsity.
Simon Willison documented hands-on exploration of the model using a DGX Spark system and Unsloth’s quantized versions, including the 72.5GB UD-IQ1_S and the 78.9GB UD-Q2_K_XL variants.
Willison noted the UD-Q2_K_XL variant with "xhigh" reasoning effort as a current favorite, sharing generated outputs in his post.
- Oct 13, 2026 · Google DeepMind — Blog
Google DeepMind releases Gemini 4 Argon with domain-leading performance and multi-domain benchmarks
Trust79 - Oct 9, 2026 · Hugging Face
Hugging Face publishes Falcon ASR, a 1.6B-parameter Arabic speech recognition model with Emirati focus
Trust84 - Oct 8, 2026 · OpenAI — News
OpenAI disrupts two AI-enabled influence operations using its models
Trust79