Mistral Launches Advanced Voice Agents For Customer Conversations
Mistral has introduced Voxtral 3, a voice agent that can understand and respond to customer inquiries in real-time, according to Search Engine Watch. This launch may enhance customer service experiences.
Original reporting · searchenginewatch.com
Original: “Mistral launches Voxtral 3 voice agents that can act during calls”
Read “Mistral launches Voxtral 3 voice agents that can act during calls” at Search Engine Watch →
How this was written. The 5:45 Ops Desk wrote this article with a language model from Search Engine Watch’s reporting (model: @cf/meta/llama-3.3-70b-instruct-fp8-fast); it describes that reporting rather than adding to it. The original, linked above, is the full story.
Key points from Search Engine Watch’s report
- Mistral has launched Voxtral 3, a voice agent that can understand and respond to customer inquiries in real-time.
- The voice agent combines speech recognition and spoken responses with tools that can act inside business systems.
- The agent's pricing varies depending on the reasoning model used, with separate transcription services available at different rates.
- Mistral suggests testing the conversation and booking record agreement after a failed action to identify discrepancies.
- The company lists API endpoints in Europe, the US, and globally, with options for running the full stack on own infrastructure or in a private cloud.
Mistral, a company specializing in AI-powered voice agents, has announced the launch of Voxtral 3, a new voice agent designed to facilitate live customer conversations. The agent combines speech recognition and spoken responses with tools that can act inside business systems. This integration may enable more efficient and personalized customer interactions.
The Voxtral 3 voice agent is built on a software development kit that supports calls and browser conversations. Mistral provides examples of how the agent can be used to check bills and book appointments while the customer is still speaking. This capability may streamline customer service processes.
The agent's architecture is designed to separate speech models from reasoning and connected tools. This arrangement allows developers to modify the reasoning model or voice without replacing the entire system, according to Mistral. The company's architecture diagram illustrates this separation.
The voice agent's pricing varies depending on the reasoning model used. Mistral charges a lower rate for its voice agent API when using a smaller reasoning model. The company also offers separate transcription services, including mini batch and real-time transcription, at different rates.
To illustrate the costs, Mistral provides calculations based on 10,000 minutes of usage. The voice agent API with a smaller reasoning model would cost $400, while the API with a larger model would cost $800. The mini batch transcription service would cost $35, and the real-time transcription service would cost $80.
For businesses considering automated appointment calls, Mistral suggests testing the conversation and booking record agreement after a failed action. This test can help identify discrepancies between the conversation and the actual reservation. The company also recommends measuring the completed booking and the caller's confirmation together to capture any discrepancies.
Mistral lists API endpoints in Europe, the US, and globally. Customers who want to run the full stack on their own infrastructure or in a private cloud can contact the company's sales team for more information. This article is based on reporting by Search Engine Watch.
Questions and answers
What is Voxtral 3?
Voxtral 3 is a voice agent designed to facilitate live customer conversations, according to Search Engine Watch.
How does the voice agent's pricing work?
The voice agent's pricing varies depending on the reasoning model used, with separate transcription services available at different rates.
What testing does Mistral recommend for automated appointment calls?
Mistral suggests testing the conversation and booking record agreement after a failed action to identify discrepancies.
