Overview

Mistral released Voxtral Transcribe 2, a new family of audio transcription models that includes both open-source and API versions. The models demonstrate real-time transcription capabilities with high accuracy for technical jargon and fast speech.

Key Facts

  • Open-source model available as 8.87GB download - developers can run transcription locally without API dependencies
  • Real-time transcription during live demo - accurately captures technical terms like Django and WebAssembly instantly
  • API model includes speaker diarization and context biasing - can distinguish multiple speakers and improve accuracy for domain-specific terms
  • Provides timestamped segments with multiple export formats - enables automated subtitle generation and searchable transcripts
  • Excellent web interface for testing transcription - makes audio content immediately accessible and searchable

Why It Matters

This represents a significant advancement in accessible AI transcription technology, offering high-quality speech-to-text without vendor lock-in through the open-source option while providing enterprise features through the API.