AI

Mistral Launches Voxtral 3 Voice Agent API and Speech Models

Tuesday, October 20, 2026Read Original

Details

  • Mistral AI announced Voxtral 3, a family of transcription, text-to-speech, and voice-agent tools for real-time conversations.
  • Voxtral Small Transcribe 3 supports multilingual transcription, speaker diarization, context biasing, and end-of-speech detection for faster agent responses.
  • Voxtral Mini TTS 3 delivers expressive speech in more than 35 languages and 50 dialects, adapting to emotion, pace, and emphasis.
  • Developers can control delivery explicitly with tags such as [emotion=sad] and [pace=slow].
  • Pro Voice enables professional voice adaptation from up to 30 minutes of recorded audio, replicating a speaker across supported languages.
  • The new SDK, built on LiveKit, combines transcription, reasoning, tool calling, and speech generation in one streamed pipeline.
  • Mistral says Voxtral 3 is available through its API, including European, US, and global endpoints, with self-deployment also supported.
  • The launch expands Mistral’s position from model provider into a broader voice-agent infrastructure competitor.

Impact

Voxtral 3 enters a market where OpenAI already offers an integrated Realtime API combining audio, reasoning, tools, interruptions, and handoffs. Mistral’s differentiation is the modular pipeline, LiveKit-based SDK, multilingual speech coverage, and European hosting or self-deployment options, which may appeal to organizations prioritizing data sovereignty and deployment control. By combining transcription, expressive synthesis, voice adaptation, and agent orchestration, Mistral could lower integration work for contact centers and hands-free applications, though adoption will depend on latency, pricing, reliability, and independent evaluations against established voice platforms.

Rift Dispatch