Voxtral transcribes at the speed of sound

2026-02-04 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Intermediate, quick

Summary

Mistral has released Voxtral Transcribe 2, a new family of two audio-to-text transcription models, following the original Voxtral released in July 2025. One model, `Voxtral-Mini-4B-Realtime-2602`, is open-weights under an Apache-2.0 license and available as an 8.87GB download from Hugging Face, offering real-time transcription capabilities. The second model, `voxtral-mini-latest`, is closed-weights and accessible via the Mistral API at a cost of $0.003 per minute, or $0.18 per hour. The Mistral API console now includes a speech-to-text playground for testing the `voxtral-mini-latest` model, providing diarized transcripts with options to download results in text, SRT, or JSON formats.

Key takeaway

For engineering teams evaluating speech-to-text solutions, consider Mistral's Voxtral Transcribe 2. Your choice between the open-weights `Voxtral-Mini-4B-Realtime-2602` and the API-based `voxtral-mini-latest` should depend on whether your priority is local deployment and cost control or leveraging advanced features like diarization and context biasing through a managed service. Test the API playground to assess its transcription quality for your specific audio data.

Key insights

Mistral's Voxtral Transcribe 2 offers both open-weights and API-based speech-to-text models with real-time and diarization features.

Principles

• Open-weights models foster broader adoption.
• API access simplifies integration for developers.

Method

The Mistral API allows audio transcription with diarization, context biasing, and timestamp granularities via a POST request to `/v1/audio/transcriptions`.

In practice

• Use `Voxtral-Mini-4B-Realtime-2602` for local, real-time transcription.
• Integrate `voxtral-mini-latest` via Mistral API for diarized transcripts.
• Utilize the API console's playground for quick testing and file uploads.

Topics

• Audio Transcription
• Mistral AI
• Open-weights Models
• Speech-to-Text API
• Real-time AI

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Mistral drops new speech-to-text AI models — Mistral's new speech-to-text models prioritize speed, privacy, and affordability for diverse enterprise applications.
• Open Weight Text-to-Speach with Voxtral TTS — Voxtral TTS offers open-weight, low-latency, multilingual voice cloning from minimal audio for diverse applications.
• New Open Audio Models 🤗 | Recap with Jeff — New open audio models from Mistral and Cohear offer efficient text-to-speech and speech-to-text capabilities.
• Mistral AI Launches Text-to-Speech Model — Mistral AI's Voxtral TTS offers open-weight, multilingual, adaptable text-to-speech for enterprise and edge devices.
• Voxtral Realtime — Voxtral Realtime delivers offline-quality ASR with sub-second latency via a natively streaming, end-to-end trained architecture.
• Introducing MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 in Microsoft Foundry — Microsoft Foundry introduces three new AI models for efficient, high-quality speech recognition, speech generation, and text-to-image capabilities.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.