Qwen3.8-Flash-Next Previews Qwen4 Architecture With 6B Active Parameters
SMRTR summary
Alibaba's Qwen team released Qwen3.8-Flash-Next on August 26, 2026, giving the public its first look at the architecture planned for Qwen4. The model has 125 billion total parameters but only activates 6 billion per token, using hybrid attention to keep costs low. This experimental, open-weight release is designed to preview a more efficient design approach Alibaba is calling "ultimate cost-efficiency."
SMRTR provides this summary for quick context. The original article belongs to Unite AI.
Read the original article