Mistral has launched Voxtral 3 voice agents that can answer callers and use connected business tools during a conversation. Its October 20 announcement lists a starting API rate of $0.04 per minute.
The release includes a LiveKit-based software development kit for calls and browser conversations. Mistral’s examples include looking up a bill and booking an appointment.
The agent connects speech with actions in business systems
Voxtral combines speech recognition, a reasoning model that can call tools, and speech generation. Developers can change the reasoning model or voice without replacing the entire pipeline, Mistral says.
The speech components handle different stages: transcription converts audio into text, while speech generation voices the response. Connected tools provide the records and actions the conversation needs.
For an appointment, that means the booking system must accept the selected slot before the agent confirms it to the caller. A useful deployment check is whether a rejected booking produces an accurate spoken response.
Pricing depends on the service and reasoning model
| Service | Published rate per minute | Cost for 10,000 minutes |
|---|---|---|
| Voice agent with Mistral Small 4 | $0.04 | $400 |
| Voice agent with a larger model | $0.08 | $800 |
| Mini batch transcription | $0.0035 | $35 |
| Mini real-time transcription | $0.008 | $80 |
The totals are calculations from Mistral’s US dollar rates. Transcription converts speech to text; the agent service also handles reasoning and responses. Telephone and implementation costs would need separate estimates.
Mistral lists European, US and global API endpoints. Organizations seeking deployment on their own infrastructure or in a private cloud are directed to its sales team.
Source: Mistral AI
Start the conversation by posting the first comment