Grok Voice 3.0
Build real-time conversational voice applications with streaming speech, configurable sessions, tools, and SIP phone integration.
Voxtral 3
mistral.ai
Mistral's speech models and voice agent SDK turn recordings and live conversations into transcripts, spoken responses and connected actions.
Visit website
Pricing
Paid
Available on
API, Web playground, WebRTC integration, Enterprise SIP integration
Category
Ai Audio
What is Voxtral 3?
Voxtral 3 is Mistral's speech model family and SDK for building transcription services, speech generation and conversational agents. Developers can process recorded audio, transcribe live streams or connect listening and speaking to tools in their applications.
It serves teams building contact-center agents, service desks and voice assistants. Hosted access uses Mistral's European, US and global API endpoints, with Studio playgrounds for evaluation. Private-cloud and on-premises deployments are arranged through Mistral.
Key features
Recorded-audio transcription
Convert files into text with speaker labels, segment timestamps and custom vocabulary to help handle domain-specific terms.
Live transcription and turn detection
The realtime transcription model processes streaming speech and detects when a speaker finishes their turn.
Controllable speech generation
Generate speech using preset or cloned voices, with emotion and pace tags to direct delivery.
Interruptible voice agents
The SDK coordinates transcription, model responses and speech playback. Users can interrupt while the agent is speaking.
Tools during conversations
Agents can search the web, query document libraries, use workspace connectors and call custom functions during a conversation.
Who is it for?
Support teams reviewing calls
Create speaker-labelled transcripts of recorded customer conversations for review.
Developers building service agents
Connect a spoken assistant to appointment or account-management functions.
Product teams adding spoken interfaces
Add live transcription and spoken replies to an application.
How to use Voxtral 3
- Create a Mistral Studio account and generate an API key.
- Choose file transcription as an initial task and prepare a sample audio file.
- Install the Python or TypeScript SDK, or use the documented direct API request.
- Send the file to the transcription endpoint and review the returned transcript.
- For conversational applications, configure a voice agent with instructions, tools and an audio interface. For SIP telephony, arrange enterprise access through your Mistral account team.
Voxtral 3 pricing
Paid
Hosted APIs have usage-based charges for transcription, speech generation and voice agents. Contact Mistral for private deployment terms.
Check current pricing ↗What to know before you use it
- SIP telephony requires enterprise access through Mistral's account team at launch. Phone sessions require server-side tools because client-side functions cannot execute in them.
- Mini Transcribe 3 shares its 65,536-token context between audio and the vocabulary list. Small Transcribe 3 is transcription-only at launch.
- Conversation reasoning and tool selection use a separate general-purpose model within the voice pipeline.
Frequently asked questions
Can Voxtral connect to telephone systems?
Yes. SIP telephony is available to enterprise customers through Mistral's account team at launch. Browser and application connections use WebRTC.
Do I need to build the surrounding application?
Yes. The SDK manages the voice pipeline; your application defines the agent's knowledge, available tools and behavior when a call ends.
Can I start with transcription alone?
Yes. The file-transcription quickstart requires an API key and an audio file, without setting up a conversational agent.