Mistral Voxtral
Mistral AI's Voxtral 3 speech-to-text and text-to-speech models plus a Voice Agents SDK built on LiveKit, from $0.04 per minute for a complete voice agent.
Voxtral 3 models in preview: Mistral calls the Voxtral 3 speech models a public preview, so expect them to change. Anyone with an API key can use them, and the Voice Agents SDK isn't labelled preview.
- Starting price
- From $0.04/min
- Setup
- Developer build
- Pricing model
- Per Minute
- Try it
- Self-serve trial
Mistral's API starts in a free mode with no card and usage limits; the docs don't say whether it covers the audio models.
At a glance
Calls & telephony
- DetailCalls
Inbound only
- Not supportedHuman handoff
No
- Not supportedPhone numbers included
No
- SupportedBring your own carrier
Yes
- SupportedSIP trunking
Yes
Enterprise-only, inbound-only SIP through your own number and carrier or PBX
Trust & access
- Not publishedHIPAA
Not published
- SupportedSOC 2
Type II
- SupportedGDPR
Yes
EU company; DPA available
- DetailLanguages
50+ languages for transcription, 35+ for speech
- SupportedFree to start
Free plan
Free API mode with no card and usage limits; audio model coverage not stated
- SupportedDeveloper API
Yes
Morak scorecard
How we score →- Ease of setupDeveloper build
Expect developers to build call logic, integrations and deployment.
- Pricing transparencyRates published
Per-minute rates are public, so you can estimate cost at your volume.
- Try before you buyFree plan
An ongoing free plan lets you test with real calls or content.
- Call & telephony controlModerate
Documented: inbound, bring-your-own carrier and SIP.
- Compliance readinessFair
Published: SOC 2 Type II and GDPR.
- Integration coverage2 of 4 covered
Connects to phone system and automation or API. Not listed: customer records (CRM, EHR or POS) and calendar.
Typical score among developer APIs
Run multilingual voice agents on Mistral's speech stack
What you get
Get transcription, reasoning and speech as one hosted pipeline on LiveKit, with a choice of model, voice and language per agent, from $0.04 a minute.
What to account for
The Voxtral 3 models are in preview, phone calls are enterprise-only and inbound-only, and agents can't transfer to a person.
Best for
- Teams wanting a European voice AI vendor
- Multilingual transcription and voice agents
- Low per-minute voice agent cost
- Swapping the model, voice or language per agent
Not a fit for
- Outbound calling, which the telephony docs rule out
- Self-serve phone agents (SIP is enterprise-only)
- Calls that must transfer to a person
What the platform provides
- Voice Agents SDK
- Hosted transcription, Mistral reasoning and Voxtral speech in one pipeline on LiveKit, for Python, Node.js and the browser.
- Speech models
- Voxtral 3 transcription in 50+ languages and text-to-speech in 35+, with voice cloning.
Connections & compatibility
Telephony
- SIP (enterprise)
AI models & voice
- LiveKit
- Mistral models
Automation & APIs
- Python SDK
- Node.js SDK
- Browser SDK
- Webhooks
Other
- Web search
Data access & permissions
What an agent built on Mistral Voxtral can touch once deployed. These are the first questions a security review will ask.
- Reads customer dataYes
- Writes to your systemsYes
- Sends messagesNo
- Triggers workflowsYes
- Records callsNo
- Needs API keysYes
- Human approval stepNo
Mistral AI's trust center lists ISO 27001, ISO 27701 and SOC 2 Type II, and it offers a DPA. Its privacy policy keeps API inputs and outputs for 30 rolling days for abuse monitoring unless zero data retention is on, but keeps Agents API data until the account closes, and the voice docs don't say which rule covers voice agent sessions. Saved voices carry a retention notice of 30 days by default. The usage policy bans using someone's voice without their prior consent, but the cloning docs don't say how consent is checked: clone only voices you have permission to use. HIPAA terms sit in a gated trust-center document. Morak has not audited these claims.
Cost at your volume
Mistral lists the voice agent API at $0.04 per minute with Mistral Small 4 for reasoning, or $0.08 per minute with a larger model. The standalone models cost $0.0035/min (Voxtral Mini Transcribe 3), $0.004/min (Voxtral Small Transcribe 3), $0.008/min (Voxtral Mini Transcribe Realtime 3) and $13 per million characters (Voxtral Mini TTS 3). These rates are on the Voxtral 3 announcement, not the general pricing page. Telephony runs through your own carrier or PBX and is set up with Mistral's enterprise team; carrier costs are separate.
Before committing
Build a browser agent from the quickstart, then test turn detection and interruptions in each of your languages at both model rates.
- Which retention rule applies to voice agent sessions, and can we turn on zero data retention?
- Can agents place outbound calls or transfer to a person, and when will SIP be self-serve?
- How does end-of-turn detection hold up on phone audio in our languages?
- What consent do you require before a voice is cloned?
How it compares
Side by side with two alternatives buyers weigh against it, on the same facts and scorecard.
| Detail | Mistral Voxtral This productMistral AI's Voxtral 3 speech-to-text and text-to-speech models plus a Voice Agents SDK built on LiveKit, from $0.04 per minute for a complete voice agent. | A different approach Open-source framework and managed cloud runtime for building and deploying realtime voice agents. Why compare: Compare LiveKit Agents if you want the same realtime transport with your own choice of speech and LLM providers. | Similar product Developer voice AI platform with a unified Voice Agent API for building real-time conversational agents. |
|---|---|---|---|
| Starting price | From $0.04/min | From $50/mo | From $4.50/hr |
| Free to start | Free plan | Free plan | Free credits |
| Calls | Inbound | Inbound & outbound | Inbound & outbound |
| HIPAA | Not published | Enterprise plans | Enterprise plans |
| Morak scorecard How we score | |||
| Ease of setup | 2 of 5, Developer build | 2 of 5, Developer build | 2 of 5, Developer build |
| Pricing transparency | 5 of 5, Rates published | 5 of 5, Rates published | 5 of 5, Rates published |
| Try before you buy | 5 of 5, Free plan | 5 of 5, Free plan | 4 of 5, Self-serve trial |
| Call & telephony control | 3 of 5, Moderate | 5 of 5, Full control | 3 of 5, Moderate |
| Compliance readiness | 3 of 5, Fair | 5 of 5, Strong | 5 of 5, Strong |
| Integration coverage | 3 of 5, 2 of 4 covered | 3 of 5, 2 of 4 covered | 3 of 5, 2 of 4 covered |