Qwen3.8-Max uses a mixture-of-experts (MoE) architecture.