R1-AQA, inspired by DeepSeek-R1, is the open-source audio question answering model from Xiaomi, Achieves SOTA performance on MMAU using reinforcement learning (GRPO).