Step-Audio-R1 is the first audio language model to successfully unlock test-time compute scaling.