Voxtral Mini Transcribe 3
$0.0035 per minuteUsage-based, per audio minuteThe 4B-parameter batch transcription model for recorded files. Official source 1 ↗ · Official source 2 ↗
Reviewed tool profile
Voxtral 3
Mistral AI's speech-to-text and text-to-speech model family with a voice agent SDK for real-time voice agents.
Sources checked
- Pricing
- Usage-based API pricing in US dollars per model: $0.0035 per audio minute for Voxtral Mini Transcribe 3, $0.004 per minute for Voxtral Small Transcribe 3, $0.008 per minute for Voxtral Mini Transcribe Realtime 3, and $13 per million characters for Voxtral TTS 3. Self-deployment is also offered, without a public price. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗ · Official source 4 ↗ · Official source 5 ↗
- Available as
- Cloud API on Mistral's platform, called with an API key from Mistral AI Studio over REST, a WebSocket for realtime transcription, or the Mistral Python and TypeScript SDKs. Mistral also offers self-deployment on your own infrastructure. Official source 1 ↗ · Official source 2 ↗
- Primary group
- Voice & Speech
Also consider
See all 6 similar tools ↓- Fish AudioVoice AI platform for expressive text-to-speech, voice cloning and transcription, with open-weight models and a pay-as-you-go API.Free plan
- DeepgramBuild speech recognition, text-to-speech, and conversational voice agents through APIs.Free trial
- Zen WhisperLocal-first Mac dictation and transcription that types into your existing apps.Free plan
What it does
Voxtral 3 is Mistral AI's speech model family, released on 20 October 2026. It has five models: three speech-to-text models (Voxtral Mini Transcribe 3, Voxtral Small Transcribe 3 and the streaming Voxtral Mini Transcribe Realtime 3) and two text-to-speech models (Voxtral TTS 3 and Voxtral TTS Realtime 3). It ships with a voice agent SDK that chains listening, reasoning with a Mistral model, tool calls and speech into one real-time pipeline. Voxtral 3 is offered through Mistral's API, and Mistral also offers self-deployment on your own infrastructure.
Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗Features
- Batch transcription with speaker labels. Voxtral Mini Transcribe 3 and the larger Voxtral Small Transcribe 3 transcribe recorded files with diarization (who spoke when), segment timestamps and custom vocabulary. Word-level timestamps are not available. Official source 1 ↗ · Official source 2 ↗
- 50 transcription languages. Mistral says the transcription models now support 50 languages, including the 24 official EU languages and many low-resource Asian and Indic languages. Official source 1 ↗
- Context biasing. You can pass key terms and free-form text to bias a transcript toward the right spelling of names and technical terms. Official source 1 ↗
- Realtime transcription with turn detection. Voxtral Mini Transcribe Realtime 3 streams text over a WebSocket with a configurable delay of 80 ms to 2,400 ms and built-in end-of-turn detection, so a voice agent knows when the caller has finished speaking. Official source 1 ↗
- Expressive text-to-speech. Voxtral TTS 3 generates speech from preset or cloned voices, with 34 inline emotion tags, slow and fast pace tags, phoneme-based custom pronunciation, streamed audio and five output formats (mp3, wav, pcm, flac, opus). Official source 1 ↗ · Official source 2 ↗
- Voice cloning and Pro Voice. Zero-shot cloning reproduces a voice from a short recording and saves it for reuse by ID; Pro Voice adaptation targets professional use and needs about 30 minutes of audio of the speaker. Official source 1 ↗ · Official source 2 ↗
- Voice agent SDK. A voice agent SDK combines realtime listening, reasoning with a Mistral model, tool calling and speech, so agents can look up information and act in business systems during live phone or web conversations. Mistral's quickstart runs it with Mistral Small and a preset voice. Official source 1 ↗ · Official source 2 ↗
Pricing & free limits
Voxtral Small Transcribe 3
$0.004 per minuteUsage-based, per audio minuteMistral's flagship, higher-accuracy transcription model for recorded files. Official source 1 ↗
Voxtral Mini Transcribe Realtime 3
$0.008 per minuteUsage-based, per audio minuteStreaming transcription over WebSocket for live audio and voice agents. Official source 1 ↗
Voxtral TTS 3
$13 per 1M charactersUsage-based, per million input charactersText-to-speech billed on the characters of input text. Official source 1 ↗ · Official source 2 ↗
Billing details
Speech-to-text models are billed per minute of audio and Voxtral TTS 3 per million characters of input text, unlike Mistral's token-priced language models. Official source 1 ↗ · Official source 2 ↗
Self-deployment on your own infrastructure is offered, but no public self-deployment price is listed. Official source 1 ↗
Free access & limits
Free access and its limits were not established by the checked official sources.
Best for & limitations
Contact-center, service-desk and virtual-assistant voice agents that hold live conversations and complete tasks in business systems. Official source 1 ↗
Transcribing recorded calls, meetings and podcast episodes with speaker labels and timestamps. Official source 1 ↗
Live captions and streaming transcription from a microphone or phone line. Official source 1 ↗ · Official source 2 ↗
Reading scripts, articles or messages aloud in a preset or cloned brand voice. Official source 1 ↗
- Public Preview. The Voxtral 3 model cards list the models as Public Preview, released on 20 October 2026. Official source 1 ↗
- Some features missing at launch. Voxtral Small Transcribe 3 handles transcription only at launch; questions or summaries about a recording use the older Voxtral 2 chat model. TTS style instructions are not available at launch, and diarization and turn detection cannot share one realtime session. Official source 1 ↗ · Official source 2 ↗ · Official source 3 ↗
- TTS language count differs by source. The launch post says Voxtral TTS supports 35+ languages and 50+ dialects, while the Voxtral TTS 3 model card says it supports 9 languages; check the docs language list for your language. Official source 1 ↗ · Official source 2 ↗
Similar tools
Tools like Voxtral 3
Published tools that share its categories.
Fish AudioVoice AI platform for expressive text-to-speech, voice cloning and transcription, with open-weight models and a pay-as-you-go API.
Visit DeepgramBuild speech recognition, text-to-speech, and conversational voice agents through APIs.
Visit Zen WhisperLocal-first Mac dictation and transcription that types into your existing apps.
Visit 
TwinMindAI memory app that transcribes your meetings and day, builds a second brain, and gives any AI your context.
Visit 
EarnoteFree, open-source lecture and meeting recorder for Mac and iPhone that transcribes on the device and writes notes, flashcards and study sheets.
Visit 
Grok Voice 3.0SpaceXSI’s speech-to-speech model for real-time voice agents and support calls, at $0.08 per minute.
Visit