The Year in AI Papers
The year/Independent research

Paper 2511.15848

Step-Audio-R1 Technical Report

Fei Tian, Xiangyu Tony Zhang, Yuxin Zhang, Haoyang Zhang, Yuxin Li, Daijiao Liu, Yayue Deng, Donghang Wu, Jun Chen, Liang Zhao, Chengyuan Yao, Hexin Liu, et al.

Published
Nov 2025
Research lab
Independent
Citations
35
GitHub
Not linked

01 In brief

Summary

Step-Audio-R1 is the first audio reasoning model to successfully benefit from deliberate thinking, addressing the historical issue where audio language models performed worse with extended reasoning.

The team identifies the root cause as 'textual surrogate reasoning,' where models reason from transcripts rather than acoustic features.

They propose Modality-Grounded Reasoning Distillation (MGRD), an iterative framework that progressively shifts reasoning from text-based to acoustic grounding through self-distillation, supervised fine-tuning, and reinforcement learning.

The model, built on Qwen2.5 32B with a frozen Qwen2 audio encoder, achieves an average score of 83.6% on speech-to-text benchmarks, outperforming Gemini 2.5 Pro (81.5%) and approaching Gemini 3 Pro (85.1%).

In speech-to-speech tasks, Step-Audio-R1 Realtime reaches a 96.1% reasoning score with 0.92s latency, surpassing GPT Realtime 0825 and Gemini 2.5 Flash Native Audio Dialog.

Key findings include that format rewards prevent reasoning collapse, moderately difficult problems are better training data than consistently failed ones, and iterative self-distillation plus DPO reduces self-cognition errors from 6.76% to 0.02%.

02 From the paper

Abstract

Recent advances in reasoning models have demonstrated remarkable success in text and vision domains through extended chain-of-thought deliberation. However, a perplexing phenomenon persists in audio language models: they consistently perform better with minimal or no reasoning, raising a fundamental question - can audio intelligence truly benefit from deliberate thinking? We introduce Step-Audio-R1, the first audio reasoning model that successfully unlocks reasoning capabilities in the audio domain. Through our proposed Modality-Grounded Reasoning Distillation (MGRD) framework, Step-Audio-R1 learns to generate audio-relevant reasoning chains that genuinely ground themselves in acoustic features rather than hallucinating disconnected deliberations. Our model exhibits strong audio reasoning capabilities, surpassing Gemini 2.5 Pro and achieving performance comparable to the state-of-the-art Gemini 3 Pro across comprehensive audio understanding and reasoning benchmarks spanning speech, environmental sounds, and music. These results demonstrate that reasoning is a transferable capability across modalities when appropriately anchored, transforming extended deliberation from a liability into a powerful asset for audio intelligence. By establishing the first successful audio reasoning model, Step-Audio-R1 opens new pathways toward building truly multimodal reasoning systems that think deeply across all sensory modalities.