StepAudio 3
Five-model audio stack for voice agents: Realtime, ASR, TTS, Gen, and Music behind one platform.
About
StepAudio 3 is StepFun's audio model family, released September 15, 2026 as five separate APIs: Realtime for full-duplex conversation, ASR for transcription, TTS for synthesis, Gen for general audio generation, and Music for interactive music creation. The family is aimed at voice agents and media generation rather than single-task speech work.
StepFun claims the Realtime model scored 98.9% on Artificial Analysis Conversational Dynamics, ranking first, and also led Speech Reasoning, while the ASR model recorded a 1.7% word error rate, tied for first on non-streaming recognition. Those are vendor-reported figures. The models are API-only on StepFun's open platform, with the Realtime model served over WebSocket and tool calling supported mid-conversation.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
- StepAudio 3 Realtime: 98.9% on Artificial Analysis Conversational Dynamics, ranked first (vendor-reported)
- StepAudio 3 ASR: 1.7% word error rate, tied first on non-streaming recognition (vendor-reported)
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (verified tier, metrics-v1). Evidence: current generation with active docs and cookbook, plus benchmark-registry integration underway. Closed weights. Release facts corroborated across StepFun's own docs and multiple outlets; docked for vendor-reported benchmark claims and preview pricing.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 6.0/10
- Maintenance · recency, releases, issue load
- 9.0/10
- Openness · license + docs
- 3.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Adoption 6: new five-branch API family on StepFun's open platform; benchmark registry integration underway
- · Maintenance 9: released Sep 15, 2026; current generation with active cookbook/docs
- · Openness 3: proprietary API-only, no downloadable weights
- · Manual 7: release facts (Sep 15 2026; Realtime/ASR/TTS/Gen/Music branches; 98.9% AA Conversational Dynamics; 1.7% WER) corroborated across StepFun docs, runtimewire, and AI Market Watch; docked for vendor-reported benchmark claims and preview pricing