Languages
This page shows how each audio capability handles languages, and which languages each Voxtral 3 model supports. Coverage differs by model and by mode, so check the model you use.
How each capability handles languages
Each capability sets or detects the language in its own way:
| Capability | How the language is set | Without a language | Details |
|---|---|---|---|
| File transcription | language, a two-letter ISO 639-1 code such as fr. Any other value returns a 422 error. | The model detects the language of the recording. | Pin the language |
| Realtime transcription | language in session.update, the same two-letter code | The model detects the language of the audio. | Session fields |
| Speech generation | Write input in the language to speak. Each preset voice has a language and an accent of its own. | Not applicable | Voices and Use a cloned voice in other languages |
| Translation | The prompt of an audio in chat request, such as "Translate this call into French" | Not applicable | Audio in chat completions |
Supported languages by model
Each model has its own list of supported languages:
| Model | Capability | Supported languages |
|---|---|---|
voxtral-mini-transcribe-3-0 | File transcription | 50 languages |
voxtral-small-transcribe-3-0 | File transcription | 50 languages |
voxtral-mini-transcribe-realtime-3-0 | Realtime transcription | 45 languages and dialects |
voxtral-mini-tts-3-0 | Speech generation | 35 languages, 50+ dialects |
TTS languages and dialects
Voxtral TTS 3 supports 35 languages with 50+ regional dialects. Language is determined by the voice, not by a request parameter: each voice is tagged with one or more languages when it is created. Write input in the language you want spoken, and pick a voice that matches.
Preset voices are available in 22 of these languages. For the rest, use voice cloning with a recording in the target language.
| Language | Code | Preset voices |
|---|---|---|
| Arabic | ar | 1 |
| Bulgarian | bg | — |
| Catalan | ca | 1 |
| Chinese | zh | 1 |
| Croatian | hr | — |
| Czech | cs | — |
| Danish | da | — |
| Dutch | nl | 1 |
| English (UK) | en_GB | 6 |
| English (US) | en_us | 4 |
| Filipino | fil | — |
| Finnish | fi | — |
| French | fr | 4 |
| German | de | 1 |
| Greek | el | — |
| Hindi | hi | 2 |
| Hungarian | hu | 1 |
| Icelandic | is | — |
| Indonesian | id | — |
| Italian | it | 1 |
| Japanese | ja | 1 |
| Korean | ko | 2 |
| Latvian | lv | — |
| Luxembourgish | lb | — |
| Norwegian | no | 1 |
| Polish | pl | 1 |
| Portuguese | pt | 1 |
| Romanian | ro | 1 |
| Russian | ru | 1 |
| Slovak | sk | — |
| Slovenian | sl | — |
| Spanish | es | 1 |
| Swedish | sv | — |
| Tamil | ta | — |
| Thai | th | — |
| Turkish | tr | 1 |
| Ukrainian | uk | 1 |
| Urdu | ur | — |
| Vietnamese | vi | 1 |
Languages without a preset voice (marked —) are still supported by the model. To use them, clone a voice from a short recording in the target language. The cloned voice is tagged with that language automatically.
Translation
Translation is a task of audio in chat completions: the model listens to the audio and writes the translation as text.