What is Step-Audio-R1?
StepFun's open audio reasoning model that thinks through speech, sounds and music before answering.
Who is Step-Audio-R1 best for?
Step-Audio-R1 is best for AI teams that want an open model able to reason over speech, sounds and music on multi-GPU servers.
Audio understanding research
It reasons over the sound itself rather than a transcript, and StepFun reports that longer reasoning makes its answers better.
Self-hosted audio Q&A
The weights and inference code are open, so you can run it on your own GPUs behind a vLLM server.
Speech, sound and music in one model
StepFun trained it across speech, environmental sounds and music, so one model covers all three.
How much does Step-Audio-R1 cost?
Step-Audio-R1 is free to download under the Apache 2.0 license, and StepFun does not sell it through its API, where only the newer step-audio-r1.5 is priced.
- 33B parameters on Hugging Face
- Runs on Linux with NVIDIA GPUs, tested on 4 data-center cards
- Inference code and a vLLM Docker image included
StepFun's API does not list Step-Audio-R1 itself. The newer step-audio-r1.5 costs $1.43 per million input tokens and $15 per million output tokens on the StepFun Open Platform. Full pricing on platform.stepfun.ai. We checked these prices on .
How do readers rate Step-Audio-R1?
No one has rated Step-Audio-R1 yet. Be the first.
Sign in to rate. One rating per account.
Is Step-Audio-R1 open source?
Yes. Step-Audio-R1's code is public under the Apache 2.0 licence, so you can run it on your own server.
On GitHub
703stars
- Last commit
- 29 Apr
- Repository age
- 12 months
- Licence
- Apache 2.0
- Self-hosted
- Yes
- Repository
- stepfun-ai/Step-Audio-R1
What are the best Step-Audio-R1 alternatives?
Step-Audio-R1 is an open source alternative to Qwen3.5-Omni and Gemini 3.5 Flash. DeepSeek V4.1 Flash and Bonsai 27B do the same job and are open source too.
Text, images, audio and video in, text or speech out, with a 256K context.
Google's stable Gemini 3.5 Flash for fast multimodal work.
DeepSeek's small, efficient open model.
PrismML's 1-bit and ternary 27B models that run locally on modest hardware.
Questions about Step-Audio-R1
What is Step-Audio-R1?
StepFun's open audio reasoning model that thinks through speech, sounds and music before answering.
Who is Step-Audio-R1 for?
Step-Audio-R1 is best for AI teams that want an open model able to reason over speech, sounds and music on multi-GPU servers.
Is Step-Audio-R1 free?
Yes. Step-Audio-R1 has a free Open weights plan: 33B parameters on Hugging Face.
Is Step-Audio-R1 open source?
Yes. Step-Audio-R1's code is public under the Apache 2.0 licence, so you can run it on your own server.
What are the best Step-Audio-R1 alternatives?
The closest alternatives on Recaply are Qwen3.5-Omni, Gemini 3.5 Flash, DeepSeek V4.1 Flash and Bonsai 27B.
Where can I try Step-Audio-R1?
On its official website, huggingface.co.
What hardware does Step-Audio-R1 need?
Step-Audio-R1 needs Linux, Python 3.10 or newer and NVIDIA GPUs with CUDA. StepFun tested it on four L40S, H100, H800 or H20 cards, served through its customized vLLM Docker image.







