Mistral releases Voxtral 3 speech models in public preview
Mistral has released four Voxtral 3 models for speech transcription and speech generation. The family includes real-time transcription, timestamped transcripts with speaker identification, and text-to-speech with voice cloning.
What changed
All four Voxtral 3 models are in Public Preview. Voxtral TTS 3 offers zero-shot voice cloning, streaming and contextual delivery. Mini Transcribe 3 and Small Transcribe 3 provide offline transcription and diarization with timestamps, while Mini Transcribe Realtime 3 provides real-time transcription.
Why it may matter for your work
These models could help teams turn recordings into timestamped, speaker-separated transcripts or add live transcription to applications. Speech generation may also help teams create spoken content from text.
What it can actually do
Generate speech with zero-shot voice cloning and streaming; transcribe audio in real time or offline; provide speaker diarization and timestamps with the offline models.
Limits to check
All four models are in Public Preview. The page links to audio pricing but does not state prices.
“We released the **Voxtral 3** audio family, our next generation of speech-to-text and text-to-speech models.”From the linked source page