Voxtral TTS 3
Audio Preview Full comparison ↗ mistral/voxtral-tts-3 · by Mistral AI
· neural-tts
Pricing — 1 offering(s)
About this offering: Mistral's text-to-speech model, in Public Preview since 2026-10-20, at $13 per million characters. It does zero-shot voice cloning and professional voice cloning, supports 35+ languages and 50+ dialects, and streams with about 90 ms to first audio (Mistral's figures). Mistral also lists a cached-input rate of $1.30 per million characters, which isn't tracked as a separate price here.
Text characters
- $13.00 / 1M characters Current 2026-10-20 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Voxtral TTS 3 fits into a cost-aware routing setup
See how →Capability profile
How Voxtral TTS 3 rates across core capability dimensions, with the task-level evidence behind each rating.
35+ languages and 50+ dialects, per Mistral's model page; vendor-reported.
Mistral's launch post shows it leading Voice Code Bench (BeSimple AI) on quality per dollar, but no independent quality ranking was found. Not rated until one exists.
Zero-shot voice cloning, plus Pro Voice adaptation from about 30 minutes of audio, per Mistral; cloning quality not independently assessed.
About 90 ms time to first audio when streaming, per Mistral's model page; vendor-reported.
Ratings are estimated — limited independent data is available for this model.
Benchmarks
| Benchmark | Score | Config | Source |
|---|---|---|---|
| Time to first audio (streaming) | 90 ms | Mistral's model page ("~90 ms time-to-first-audio"); vendor-reported, measurement setup not stated. | source ↗ |
Operator guidance
Choose Voxtral TTS 3 for low-cost streaming TTS with voice cloning across many languages: at $13 per 1M characters it's among the cheapest per-character TTS Modelglass tracks. It's a Public Preview and its quality claims are vendor-reported, so test it against Inworld TTS-2 Flash or Fish Audio S2.1 Pro at a similar price.
Use cases
- Low-cost expressive speech for voice agents and assistants, at $13 per 1M characters
- Multilingual and dialect-specific speech in a customer's own cloned voice
Limitations
- Public Preview: Mistral allows silent updates and gives no guaranteed path to general availability
- Latency, language and quality figures are vendor-reported
- No independent quality ranking yet