---
canonical: "https://www.softatlas.io/mistral/products/voxtral"
name: "Voxtral"
provider: "Mistral"
category: "Artificial Intelligence / Generative Media & Voice AI / Speech-to-Text APIs"
review_count: 0
format: "CORE_PRODUCT"
ai_powered: true
launched: 2025
website: "https://mistral.ai/solutions/speech"
updated: "2026-10-23"
---

# Voxtral

By [Mistral](/mistral.md)

Voxtral is Mistral's family of speech models, served through the Mistral API. It covers batch transcription, realtime transcription with a latency set per session, and text-to-speech with preset or cloned voices. Transcription identifies speakers, returns segment timestamps, accepts a custom vocabulary and detects the spoken language. Voxtral 3, the current generation, is in public preview since October 2026; the first models were released with open weights in July 2025.

## Features
- Transcribe recordings with Voxtral Mini Transcribe, or Voxtral Small Transcribe for higher accuracy
- Transcribe live audio with a streaming delay set per session, from 80 ms
- Identify speakers, on recordings and in realtime
- Bias recognition toward a custom vocabulary of names and jargon
- Detect the spoken language automatically, or set it
- Generate speech from text with preset or cloned voices, in batch or streaming
- Summarize or ask questions about a recording through audio in chat

## Built for
DEVELOPERS, CUSTOMER_SUPPORT, OPERATIONS

Source: https://www.softatlas.io/mistral/products/voxtral