Research2h ago
Research
Qwen Releases Flash-Next MoE Model Preview
The Qwen3.8-Flash-Next model utilizes a Mixture-of-Experts architecture with 125B total parameters but only 6B active. This release serves as an early architectural preview for the upcoming Qwen4. Developers can currently run quantized versions via Unsloth. The design prioritizes inference speed without sacrificing the reasoning capabilities required for complex multimodal tasks.