Tasks and inputs
Voxtral TTS 3 is listed for Text → Speech, Text-to-speech. It accepts text and produces audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Voxtral TTS 3 is Mistral's speech synthesis model for batch and streaming audio, with voice cloning, emotion tags and 35+ languages at $13 per 1M chars.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $13 per 1M characters. Confirm on your own Mistral account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Three names, one model. Mistral's launch post calls it Voxtral Mini TTS 3, the model table says Voxtral TTS 3, and at least one product card reportedly reads Voxtral Speech 3. The API ID settles it: voxtral-mini-tts-3-0. Use that string in code and the naming question goes away.
What you get is a single text-to-speech model that handles both offline batch generation and realtime streaming, released on 20 October 2026. It replaces the March 2026 Voxtral TTS (voxtral-mini-tts-2603), deprecated on 31 October.
There are three routes to a voice:
Cloning a real person's voice needs that person's informed consent. Build the consent record into your onboarding flow before the feature ships, not after.
Beyond plain text, the model takes instructions. There are 34 emotion tags and 2 pace tags, plus support for emphasis and contextual prompting, and the demos show phoneme hints for words it would otherwise mispronounce. Audio comes out in five formats, 24 kHz WAV among them.
| Item | Detail |
|---|---|
| API ID | voxtral-mini-tts-3-0 |
| Modes | Batch and realtime streaming |
| Languages | 35+ languages, 50+ dialects |
| Preset voices | 18 languages |
| Output | 5 formats, including 24 kHz WAV |
| Expressive controls | 34 emotion tags, 2 pace tags |
| Indicative price | $13 per 1M characters |
| Weights | Closed |
Character billing makes estimates easy. A 10,000-character article read aloud comes to about 13 cents. A support line speaking 200 characters per response, 50,000 times a day, uses 10 million characters: about $130 a day at the indicative rate. Confirm the price on your own Mistral account before modelling anything serious.
Its strongest argument is the pairing: the same vendor's realtime recogniser on the input side, this model on the output, and Mistral's LiveKit-based voice agent API if you want the whole loop managed. If you only need narration, compare on voice quality alone.
Listen, don't just score. Have three people rate twenty sentences from your real content, including numbers, addresses and brand names, across two or three preset voices and one cloned voice. Note mispronunciations, then check whether a phoneme hint fixes them. For streaming, measure time to first audio from your region, since the API runs from EU, US and global endpoints.
Other speech models are indexed under text to speech, and the rest of the family is on the Mistral provider page.
Compare the model on the work you need to complete.
Voxtral TTS 3 is listed for Text → Speech, Text-to-speech. It accepts text and produces audio. Use the publisher’s input format and preprocessing for your chosen deployment.
Compare pronunciation, voice consistency and time to the first playable audio. Check the service’s billing unit for the actual script.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
zyphra
ZONOS3-Mini is Zyphra's open multilingual text-to-speech model: 800M MoE with ~100M active, zero-shot voice cloning in 43 languages and 80 ms first audio.
Alibaba
Qwen-Audio-3.1-TTS Plus is Alibaba's premium text-to-speech model for brand voices and dubbing, with fine-grained emotion tags and noise-robust voice cloning.
fish-audio
Fish Audio Drama 3 is an expressive 35B MoE text-to-speech model steered by bracketed directions, covering 83 languages, with downloadable weights.
heygen
HeyGen Voice is HeyGen's default text-to-speech model: about 39 languages, public and cloned voices, streaming with character timestamps.
heygen
HeyGen Voice Turbo is the low-latency version of HeyGen's TTS model for public voices and instant clones, at $4 per 1M characters.
Alibaba
Qwen-Audio-3.1-TTS Flash is Alibaba's streaming text-to-speech model for voice assistants, live chat and customer service, with system and cloned voices.