As of: 2026-10-25 23:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/voxtral-mini-tts-3/

# Voxtral TTS 3 (Voxtral Mini TTS 3)

Mistral AI, Voxtral family. Status: preview. Type: speech and audio. Released: 2026-10-20.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| Mistral API | `voxtral-mini-tts-3-0` | https://api.mistral.ai/v1/audio/speech | https://docs.mistral.ai/models/voxtral-mini-tts-3-0 |
| Mistral Studio docs (text-to-speech) |  |  | https://docs.mistral.ai/studio/audio/text-to-speech/overview |

## Pricing

- Per 1M characters: $13
- Unit as published: USD per 1M characters (Mistral API)
- Source: https://docs.mistral.ai/models/voxtral-mini-tts-3-0
- Last checked: 2026-10-25

## Specs

- Input: text, audio
- Output: audio
- Context window: not published
- Max output: not published
- Open weights: no
- Licence: proprietary

## Where it stands

- No newer speech and audio model in the registry yet.
- Before it from Mistral AI: Voxtral TTS (2026-03-23) https://postcutoff.com/m/voxtral-tts/
- Closed weights; access is hosted only.

## What stands out

- Context-following emotion with explicit controls: Adapts emotion, pacing and emphasis to the conversation; explicit `emotion`, `pacing` and `emphasis` instruction variables; the model card lists 34 emotion tags and 2 pace tags. Source: https://mistral.ai/news/voxtral-3
- Pro Voice adaptation: Professional voice cloning reproduces a speaker's character, emotional range and style across languages from about 30 minutes of audio; zero-shot cloning also listed on the card. Source: https://mistral.ai/news/voxtral-3
- Streaming with ~90 ms time to first audio: Batch and realtime synthesis on /v1/audio/speech, roughly 90 ms time-to-first-audio and five output formats (model card). Source: https://docs.mistral.ai/models/voxtral-mini-tts-3-0

## Notes

Artificial Analysis TTS arena (https://artificialanalysis.ai/text-to-speech/leaderboard), read 2026-10-25: rank #12, Elo 1219, $13.0 per 1M characters. Docs: v3.0, 'Public Preview', 'Premier' (API-only; no open weights found, unlike the CC BY-NC Voxtral TTS of March 2026). Language count conflicts: the launch post says 35+ languages and 50+ dialects (voice presets in 18 languages), while the model card says nine supported languages; unresolved. Parameter count not stated. Voice-agent SDK (LiveKit-based, SIP/WebRTC) launched alongside at $0.04/min (Mistral Small 4) or $0.08/min (larger model). Successor of voxtral-tts (voxtral-tts-2603).

Mistral's Voxtral 3 text-to-speech model for voice agents and narration.

Call `POST https://api.mistral.ai/v1/audio/speech` with `model: voxtral-mini-tts-3-0` (request schema and voices: https://docs.mistral.ai/studio/audio/text-to-speech/overview).

Sources: https://mistral.ai/news/voxtral-3 · https://docs.mistral.ai/models/voxtral-mini-tts-3-0 · https://docs.mistral.ai/models/overview · https://artificialanalysis.ai/text-to-speech/leaderboard

Model page: https://postcutoff.com/m/voxtral-mini-tts-3/