China’s “Flash” Model Push: Qwen and DeepSeek Target Faster, Cheaper Inference