Qwen3 TTS API
Available NewQwen3 TTS is Alibaba's open text-to-speech model, served here by our second upstream provider on the OpenAI-compatible speech endpoint. On our 2026-11-18 check all nine preset voices answered, an "instructions" field changed the delivery, a German sentence came back as German (we transcribed it to confirm), and a one-sentence request took about 2.5 seconds as MP3 or WAV. The host documents voice cloning from a short sample through a voice_id parameter and streaming PCM output; we pass both through but have not tested them.
Quickstart
Set INFERENCE_API_KEY to your key from the API Keys page, then run:
curl https://api.inferenceapis.com/v1/audio/speech \
-H "Authorization: Bearer $INFERENCE_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen/Qwen3-TTS",
"input": "Hello! This is Qwen3 TTS speaking.",
"voice": "Vivian",
"response_format": "mp3"
}' --output speech.mp3
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.inferenceapis.com/v1",
api_key=os.environ["INFERENCE_API_KEY"],
)
with client.audio.speech.with_streaming_response.create(
model="Qwen/Qwen3-TTS",
voice="Vivian",
input="Hello! This is Qwen3 TTS speaking.",
response_format="mp3",
) as response:
response.stream_to_file("speech.mp3")
import fs from "node:fs";
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.inferenceapis.com/v1",
apiKey: process.env.INFERENCE_API_KEY,
});
const audio = await client.audio.speech.create({
model: "Qwen/Qwen3-TTS",
voice: "Vivian",
input: "Hello! This is Qwen3 TTS speaking.",
response_format: "mp3",
});
fs.writeFileSync("speech.mp3", Buffer.from(await audio.arrayBuffer()));
When to use it
- You want to steer tone and emotion with plain-language instructions ("speak slowly and calmly")
- Ten languages from one model, including Korean, Russian and German, which Kokoro does not cover
- Voice cloning from a short sample, which the host documents through a voice_id parameter
- Price per character matters most: Kokoro is five times cheaper
- You need English voice variety: Kokoro has 54 voices, this model has 9
What it costs in practice
| Workload | Cost |
|---|---|
| One 1,000-word article (about 6,000 characters) | $0.16 |
| One hour of narration (about 54,000 characters) | $1.40 |
| One million characters (about 18 hours of speech) | $26.00 |
Computed from the live rate below. There is no subscription or minimum; new accounts start with free credit.
Endpoint
Content-Type: application/json
/openai/v1/… is accepted too, so Groq-style base URLs work with only the host changed.
Model IDs and aliases
Send any of these as model; they all resolve to this model and bill at its rate.
| ID | Note |
|---|---|
Qwen/Qwen3-TTS | Canonical |
qwen3-tts | Alias |
Qwen/Qwen3-TTS-12Hz-1.7B-CustomVoice | Alias |
Request body
| Parameter | Type | Description |
|---|---|---|
model required |
string | Model ID: Qwen/Qwen3-TTS |
input required |
string | The text to speak. Up to 20,000 characters per request. |
voice |
string | Voice id. Default Vivian. See the voice list below. |
response_format |
string | mp3 (default), wav or raw (PCM). |
speed |
number | Playback speed multiplier, e.g. 1.1. |
language |
string | Language hint for multilingual voices, e.g. en, ja. |
Voices
Click a voice to copy its id. Hear them in the playground.
Response
Binary audio (Content-Type: audio/mpeg for mp3, audio/wav for wav).
Headers:
X-Usage-Characters: 42 # characters billed for this request
Pricing
| Price | $26.00 / 1M characters |
| Billing | Pay as you go from prepaid credits. No subscription, no daily request or token cap. See all pricing. |
Notes
- Voice ids are case-sensitive names: Vivian, Serena, Uncle_Fu, Dylan, Eric, Ryan, Aiden, Ono_Anna and Sohee. OpenAI's voice names (alloy, nova, onyx and the rest) are mapped to the closest of these, so code written for OpenAI's speech endpoint runs unchanged.
- Add an "instructions" string to the request to set tone, pace or emotion; it is passed straight to the model.
- Languages: English, Chinese, Japanese, Korean, German, French, Russian, Spanish, Italian and Portuguese. Pick a voice and write the text in the language; there is no language parameter to set.
- $26 per million characters: more than Kokoro ($5.20) and Orpheus ($19.50). Use it for the languages and the instruction control, not for bulk English narration.
Compare Qwen3 TTS with…
Languages
Per-language voice lists and samples: text-to-speech languages.
Errors
Errors use the OpenAI envelope: {"error": {"message", "type", "code"}}. 401 missing or invalid key · 402 insufficient_balance · 404 model_not_found · 503 model_unavailable · 502 backend error, safe to retry. Full table with what to retry: API documentation. Errors from other providers: provider error reference.
