Alibaba's Qwen3.8-Flash Rewrites the AI Cost Curve, Previews Qwen4 Architecture
Alibaba Group on Aug. 26 released Qwen3.8-Flash, a multimodal mixture-of-experts model that activates only 6 billion of its 125 billion parameters per token—delivering frontier-level performance at a price point that undercuts every major competitor and, crucially, doubles as the architectural blueprint for the forthcoming