StepFun
Step Audio 2
StepFun’s end-to-end speech model — TTS, STT and voice chat in a single checkpoint.
Intelligence
89.0 #22
AAQI 89.0
Output Speed
150 t/s
ITL 6.7 ms/tok
Blended price (3:1)
$0.015 /1M
In $0.012/M · Out $0.024/M
Context
64K
≈ 160 A4 pages · max out 8,192
Quality Domain Radar
6-Axis Frontier Evaluation vs Category Average & Flagship
Step Audio 2
Claude Mythos Preview (#1)
Average (68)
Radar axes are approximated from category scores where direct benchmark measurements are missing — see Methodology.
Strengths & Trade-offs
Where it wins
- Single model replaces a whole voice pipeline
- Best-in-class Chinese speech quality
- Understands tone and emotion, not just words
Trade-offs to watch
- English prosody below Eleven/Cartesia tiers
- Ecosystem concentrated in APAC
- No self-hosting
Killer Use Cases
Speech-to-speech agents
Direct voice chat without TTS/STT glue.
Chinese voice products
Native Mandarin/Cantonese synthesis and ASR.
Emotion-aware bots
Reacts to caller sentiment in real time.
Developer Quickstart & API
Set
STEPFUN_API_KEY first — get an API keycurl https://api.stepfun.com/v1/chat/completions \
-H "Authorization: Bearer $STEPFUN_API_KEY" \
-H "content-type: application/json" \
-d '{
"model": "step-audio-2",
"messages": [{"role": "user", "content": "Refactor this function for O(n) time and explain the change."}]
}'import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.stepfun.com/v1",
api_key=os.environ["STEPFUN_API_KEY"],
)
resp = client.chat.completions.create(
model="step-audio-2",
messages=[{"role": "user", "content": "Refactor this function for O(n) time and explain the change."}],
)
print(resp.choices[0].message.content)const res = await fetch("https://api.stepfun.com/v1/chat/completions", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.STEPFUN_API_KEY}`,
"content-type": "application/json",
},
body: JSON.stringify({
model: "step-audio-2",
messages: [{ role: "user", content: "Refactor this function for O(n) time and explain the change." }],
}),
});
console.log((await res.json()).choices[0].message.content);Official Resources & Links
API deployment
Compare hosts & latency
Throughput, TTFT and cache pricing across Groq, Cerebras, Together, DeepInfra, Fireworks, Bedrock and Azure.
GroqCerebrasTogether AIDeepInfraFireworks AIAWS BedrockAzure AI
Scores by Source
Category Scores
Benchmark Results
Head-to-head Live Matchups
Live Specs & Architecture
Under the hood
Step Audio 2 — unified speech model
- One checkpoint for TTS, STT and speech-to-speech chat
- Paralinguistic understanding — emotion, emphasis, age cues
- Voice-to-voice tool calling for agent stacks
- Strong Mandarin/Cantonese coverage
Release date
Sep 24, 2026
Architecture
Step Audio 2 — unified speech model
Parameters
Not disclosed
License
Proprietary
Context Window
64,000 tokens
Max Output
8,192 tokens
Modalities
text + audio
Cache Pricing
read $0.004 · write $0.015 /1M
Reasoning
—
Input Price
$0.012/M
Output Price
$0.024/M
Output Speed
150 t/s
Time to First Token
—
Data Updated
Oct 22, 2026
Estimated monthly bill
Estimated monthly spend
$0.00
Calculating savings vs GPT-4o...
Cite & embed
GitHub Markdown badge
Loading badge markdown...BibTeX
@misc{llmpodium2026stepaudio2,
title = {Benchmark Analysis and Intelligence Evaluation of Step Audio 2},
author = {LLMPodium Research},
year = {2026},
url = {https://llmpodium.com/models/step-audio-2}
}Embed Score Badge
Add a live Podium Score badge to your GitHub or Hugging Face README — it links back to the live ranking.
Preview
Markdown
Loading badge snippet...HTML
Loading badge snippet...More from StepFun
Alternative Models
Closest alternatives to Step Audio 2 based on capabilities, pricing, and benchmark scores.