About this model
High-quality multilingual text-to-speech with voice cloning and ultra-low latency. Qwen3-TTS 1.7B Base is the foundation text-to-speech model from Alibaba's Qwen3-TTS family.
Pricing
| Rate | RODI | USD (ref.) | Unit |
|---|---|---|---|
| Per unit | |||
| Per 1K characters | 11.4 | ~ 0.015 | per 1K characters (RODI/1K chars, USD/1K chars) |
RODI prices include RodiumAi markup and upstream fees. USD figures are wholesale reference rates.
Providers & pricing
| Provider | Input | Output | Cached | Free tier | Status |
|---|---|---|---|---|---|
 Runware Default All rates (1)
| - | - | - | - | No data |
Capabilities
Streaming
Tool calling
Vision
JSON mode
Reasoning
API usage
Synthesize speech with client.speech(): binary POST /v1/audio/speech.
Examples
…RodiumAi SDK
speech() returns raw audio bytes (save to disk or stream to clients).
…Use cases
- Voice assistants, IVR menus and phone bots.
- Narration of articles, newsletters and product tours.
- Audio versions of notifications, receipts and alerts.
- Voice-overs for videos, slides and e-learning.
- Accessibility: read screen content aloud.
Notes
- Endpoint:
POST https://api.rodiumai.io/v1/audio/speech, the OpenAI Speech format. The response body is the audio file, not JSON: write the bytes to a file. inputandmodelare required.voiceis optional. The OpenAI voice names (alloy,echo,nova, ...) are accepted for every model and are mapped to a comparable native voice when the model has its own voice library. Pass a native voice name to choose it precisely.response_formatacceptsmp3,wav,flac,opusandpcmwhere the model supports them. Other values fall back to the model default.speedis forwarded to models that support it.- Billing follows the pricing table above (per character or per token depending on the model). Keep each request to a few paragraphs and split long texts.