Voxtral 3: Mistral Voxtral Pricing & Models

Voxtral 3 is Mistral’s speech family: transcription from $0.0035/min, text to speech at $13 per 1M characters and a voice agent API. See prices and limits.

Mistral released Voxtral 3 on October 20, 2026. It’s three speech-to-text models, a text-to-speech model that runs in batch or realtime, and a voice agent SDK that chains listening, reasoning, tool calls and speech into one live pipeline. If you searched for “mistral voxtral”, Voxtral 3 is the current version. We checked every price on Mistral’s API pricing page and the Voxtral 3 announcement on October 24, 2026.

Is Voxtral 3 free or open source?

No. Voxtral 3 is a paid API in public preview, billed per minute of audio or per character of speech. Mistral hasn’t published Voxtral 3 weights: the newest Voxtral checkpoints on Mistral’s Hugging Face page are older models, such as Voxtral Mini 4B Realtime and Voxtral 4B TTS. Mistral says companies can run the full stack on-premises or in a private cloud, but that means contacting its sales team.

That’s a change from earlier Voxtral releases, which had downloadable weights. If you need a free model you can run yourself, look at open TTS models instead.

Voxtral 3 Pricing (2026)

Model	What it does	Price
Voxtral Mini Transcribe 3	Batch transcription of recorded files (4B)	$0.0035 per minute
Voxtral Small Transcribe 3	Flagship transcription with speaker labels	$0.004 per minute
Voxtral Mini Transcribe Realtime 3	Live transcription under 500 ms	$0.008 per minute
Voxtral TTS 3	Text to speech, batch or realtime	$13 per 1M characters
Voice agent API with Mistral Small 4	Listen, think and speak in one pipeline	$0.04 per minute
Voice agent API with a larger model	Same, with a stronger reasoning model	$0.08 per minute

Some quick maths. An hour of recorded audio costs about $0.21 with Mini Transcribe 3 and $0.24 with Small Transcribe 3. A 10,000-word script is roughly 60,000 characters, which costs about $0.78 to voice with TTS 3. A 10-minute support call through the voice agent API costs $0.40 with Mistral Small 4.

Text to speech also got cheaper: the older Voxtral TTS is listed at $16 per 1M characters, so TTS 3 is about 19% less. Live transcription went the other way, from $0.006 to $0.008 per minute. For context, compare these rates with ElevenLabs, which bills by plan credits instead of per minute.

What’s new in Voxtral 3

• 50+ transcription languages: including the 24 official EU languages and more Asian and Indic languages.
• Speaker labels and custom terms: diarization shows who said what, and context biasing lets you add product names or jargon without fine-tuning.
• Turn detection: the realtime model predicts when someone has finished speaking, so agents reply without awkward gaps.
• Expressive speech: TTS 3 adjusts emotion, pace and emphasis to the conversation, in 35+ languages and 50+ dialects.
• Pro Voice: a professional voice adaptation that copies a speaker’s character and range from about 30 minutes of audio.
• Voice agent SDK: agents can call tools, search the web and use your own data, and connect over SIP phone lines or WebRTC in browsers and apps.

Mistral claims Small Transcribe 3 beats Whisper large-v3, ElevenLabs Scribe v2 and Gemini 3.5 Transcribe on multilingual word error rate. Those are Mistral’s own tests, so try it on your own audio. If you’re comparing realtime voice stacks, our Eleven Realtime vs GPT Realtime breakdown covers the main rivals.

How to try Voxtral 3

1. Create an account in Mistral Studio and add an API key.
2. Call the transcription models by ID, for example voxtral-small-transcribe-3, or pick Voxtral TTS 3 for speech.
3. Choose the European, US or global endpoint, depending on where your audio may be processed.
4. For phone agents, use the voice agent SDK and connect it to your SIP or contact-centre system.

Limitations

• Public preview: model behaviour and prices can still change.
• No open weights: you can’t download Voxtral 3; self-hosting needs a deal with Mistral.
• Realtime costs more: live transcription is more than twice the batch rate.
• Voice cloning needs a lot of audio: about 30 minutes, not a few seconds.

Verdict

Voxtral 3 is one of the cheapest ways to add transcription and natural speech to an app, and the EU-hosted option matters if your data can’t leave Europe. Developers who need a free, self-hosted voice should look at Ming-3.0 TTS, and for more voice options see our ElevenLabs alternatives.

Voxtral 3 gives developers low per-minute prices for transcription and a full voice agent pipeline from one EU provider.

Use it for transcripts and phone agents; pick an open model instead if you need weights you can run yourself.