What is Voxtral TTS?
Voxtral TTS is Mistral's first text-to-speech model, a 4B-parameter model built on Ministral 3B. It reads text aloud in 9 languages, either in a preset voice or in one cloned from a short sample, and copies the sample's accent, rhythm and emotion instead of taking style tags. A French voice reading English keeps its French accent, which suits speech-to-speech translation with Voxtral Transcribe.
It is Mistral's current speech model. The open weights come with 20 preset voices under a non-commercial licence, and Mistral points voice cloning to its API and Studio.
Who is Voxtral TTS best for?
Voxtral TTS is best for developers building voice agents, support lines and multilingual voiceovers who want cloned voices in 9 languages at a low price per character.
Voice agents and support lines
Mistral measures 70 ms of model latency on a typical 500-character request, and streamed pcm audio starts in about 0.8 seconds end to end.
One brand voice in many languages
A voice cloned from a few seconds of audio can speak all 9 supported languages and keeps the speaker's accent.
Speech-to-speech translation
Paired with Voxtral Transcribe, it turns speech in one language into speech in another in the original speaker's voice.
Research and non-commercial projects
The open weights run on a single GPU with 16GB or more at no cost, with 20 preset voices, as long as the project is not commercial.
What's included with Voxtral TTS?
Voice cloning from a short clip, saved voices, cross-lingual voices, streaming output, five audio formats and long passages through the API.
- Voice cloning from a short clip
- A few seconds of reference audio is enough, and the model follows the sample's intonation, rhythm and emotion without prosody or emotion tags.
- Saved voices
- The Voices API stores a cloned voice with a name, languages and tags, so later requests only pass its ID.
- Cross-lingual voices
- A voice recorded in one language can read text in another of the 9 languages, and mixed-language text works too.
- Streaming output
- Audio streams as it is made, starting in about 0.8 seconds end to end with pcm output or about 3 seconds with mp3.
- Five output formats
- The API returns mp3, wav, pcm, flac or opus; self-hosted, the model can also write AAC.
- Long passages
- The model makes up to two minutes of audio at a time, and Mistral's API joins longer pieces for you.
How much does Voxtral TTS cost?
Voxtral TTS costs $0.016 per 1,000 characters on Mistral's API, or $0.008 in batch mode, and its open weights are free to run yourself for non-commercial use only.
- CC BY-NC 4.0 licence, non-commercial use only
- 20 preset voices on Hugging Face; voice cloning is on the API
- Serve it yourself with vLLM-Omni on a 16GB GPU
- Half the standard API rate
- For high-volume jobs that can wait
- Preset voices or a voice cloned from a few seconds of audio
- 9 languages with streaming output
- Try it first in the Mistral Studio playground
- Priority inference for predictable, fast responses
Mistral's Free plan includes $10 of API credits a month, which covers about 625,000 characters at the standard rate, and Pro ($14.99 a month) includes $30; plan credits are used before pay-as-you-go billing. The open weights' licence rules out commercial use, so commercial products use the API or a separate licence from Mistral. Prices are in US dollars; the page also lists euro prices. Full pricing on mistral.ai. We checked these prices on .
How do readers rate Voxtral TTS?
No one has rated Voxtral TTS yet. Be the first.
Sign in to rate. One rating per account.
Voxtral TTS pros and cons
It sounds natural to many testers and costs far less than ElevenLabs, but cloning is API only, the open weights are non-commercial and it speaks just 9 languages.
Pros
- Many Reddit testers found the voices clean and natural, and people who tried cloning in Mistral's hosted demos called the results very good.
- At $0.016 per 1,000 characters it costs 40% of ElevenLabs Flash and a fifth of Eleven v3 on their APIs, and batch jobs halve that.
- In Mistral's own listening test, 68.4% of ratings favoured its cloned voices over ElevenLabs Flash v2.5, at similar time to first audio.
Cons
- Voice cloning works only through Mistral's API and Studio: the open weights ship with preset voices, and users found Mistral left out the part that turns a sample into a voice.
- The open weights are CC BY-NC 4.0, so a commercial product has to use the paid API or a separate licence from Mistral.
- Only 9 languages, with no Chinese, Japanese, Korean, Russian or other Slavic languages; several users passed on it for that reason.
- Quality reviews are mixed: some users heard robotic stretches or an unnatural rhythm, and delivery can be steered only through the voice sample, not emotion tags.
What are the best Voxtral TTS alternatives?
Three tools on Recaply do Voxtral TTS's job. Chatterbox Turbo is open source, so you can run it on your own server.
Eleven v3FreeChoose it when you need 70+ languages or multi-speaker dialogue with more expressive delivery; it costs $0.08 per 1,000 characters on ElevenLabs' API, five times Voxtral's rate.
Sonic 3.6FreeChoose it when your voice agent needs more languages: Cartesia covers 44, clones from 10 seconds of audio, and its $5-a-month Pro plan includes a commercial licence.
Chatterbox TurboOpen SourceChoose it when you need open weights you can clone voices with and ship commercially; it is MIT licensed and needs about 5 seconds of audio, but speaks English only.
Questions about Voxtral TTS
What is Voxtral TTS?
Voxtral TTS is Mistral AI's text-to-speech model for voice agents and multilingual voiceovers, cloning a voice from a few seconds of audio and speaking 9 languages through Mistral's API.
Who is Voxtral TTS for?
Voxtral TTS is best for developers building voice agents, support lines and multilingual voiceovers who want cloned voices in 9 languages at a low price per character.
Is Voxtral TTS free?
Only in part. Voxtral TTS's open weights are free to download and run yourself, but the CC BY-NC 4.0 licence allows non-commercial use only. The hosted API costs $0.016 per 1,000 characters, and Mistral's Free plan includes $10 of API credits a month.
How much does Voxtral TTS cost?
Voxtral TTS's paid plans are Batch at $0.008 per 1,000 characters, Standard at $0.016 per 1,000 characters and Priority at $0.028 per 1,000 characters.
Is there an open source alternative to Voxtral TTS?
Yes. Chatterbox Turbo is an open source tool that does the same job, so you can run your own.
What are the best Voxtral TTS alternatives?
The closest alternatives on Recaply are Eleven v3, Sonic 3.6 and Chatterbox Turbo.
Where can I try Voxtral TTS?
On its official website, mistral.ai.
Which languages does Voxtral TTS speak?
Voxtral TTS speaks 9 languages: English, French, German, Spanish, Dutch, Portuguese, Italian, Hindi and Arabic. It can also use a voice recorded in one language to speak another, so a French speaker's voice can read English text with a French accent.
Can I use Voxtral TTS commercially?
Yes, through Mistral's API, which bills $0.016 per 1,000 characters. The open weights on Hugging Face are CC BY-NC 4.0, which rules out commercial use, and Mistral says commercial self-hosting of its open-weight models needs a separate Mistral licence.
Can the Voxtral TTS open weights clone voices?
Not out of the box. Mistral's Hugging Face release comes with 20 preset voices and sends voice customisation to Mistral Studio, and users report they could not clone voices locally. Cloning from a short sample works through the API and Studio.
Is Voxtral TTS better than ElevenLabs?
On cost, clearly: $0.016 per 1,000 characters against $0.04 for ElevenLabs Flash and $0.08 for Eleven v3. On sound, Mistral's own test favoured Voxtral over Flash v2.5 and rated it level with v3, but ElevenLabs covers 70+ languages to Voxtral's 9.
Our verdict
Voxtral TTS is a low-cost, natural-sounding choice for voice agents and cloned brand voices in its 9 languages, as long as you use Mistral's API.
It costs a fraction of ElevenLabs per character, and testers liked its sound and its cloning from a few seconds of audio.
- Skip it for
- Self-hosted or commercial open-weight projects that need cloning, and anyone working outside the 9 languages.
- Check first
- Clone your own voice in the Mistral Studio playground and listen in every language you plan to use.
Voxtral Transcribe 2
Voxtral 3
Eleven v4
Qwen3-TTS
MAI-Voice-2
Gemini 3.8 Flash TTS