Xiaohongshu FireRed launches universal audio language model FireRedAudio

The Xiaohongshu FireRed team has launched the general-purpose audio language model FireRedAudio, based on the 9B parameter Qwen3.5 architecture and employing a decoupled continuous representation design. The model simultaneously supports six tasks: ASR, audio question answering, 1-hour long audio comprehension, zero-shot TTS, instruction TTS, and speech editing. The model has achieved leading results in benchmarks such as MMAU, MMSU, 102-language ASR, and Instruction TTS, with an average error rate of only 14.94% in FLEURS-102.