The AI Colony
Posts
Mistral releases Voxtral, its first open source AI audio model

Mistral releases Voxtral, its first open source AI audio model

Midweek AI Digest

The AI Colony
July 16, 2025

In partnership with

Hello Techies,

Welcome to our Wednesday "Bit-Size Update" on how the week started in the AI Industry.

In today's menu:

AI quote of the day
Tip of the Day
Featured Product
Bit-Size AI update
Top products this week

❝

The standard you walk past is the standard you accept

David Hurley

Tip of the Day

Start learning AI in 2025

Everyone talks about AI, but no one has the time to learn it. So, we found the easiest way to learn AI in as little time as possible: The Rundown AI.

It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.

Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses – tailored to your needs.

FEATURED PRODUCT

An AI-powered platform that features various user tools, including AI-powered tools to help users create marketing materials.

- The AI Library.

Bit-Size AI Update

Mistral releases Voxtral, its first open source AI audio model

French AI innovator Mistral has debuted Voxtral, its first open-source audio model family, designed to redefine how developers build voice-enabled applications. Until now, teams faced a frustrating choice: opt for open-source models with poor accuracy or rely on expensive, proprietary services that lock them into specific platforms. Mistral claims Voxtral solves this dilemma by offering “state-of-the-art accuracy and native semantic understanding in the open, at less than half the price of comparable APIs.”

Voxtral comes in two main tiers: Voxtral Small, a 24-billion-parameter model optimized for full-scale business deployments, and Voxtral Mini, a lighter 3B model ideal for local or edge processing. There’s also a dedicated transcription-only version called Voxtral Mini Transcribe, which rivals OpenAI’s Whisper, at a fraction of the cost, reportedly under $0.001 per minute.

The models boast long-context capabilities (up to 30 minutes for transcription and 40 minutes for understanding), as well as multilingual performance across English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian. Powered by Mistral’s Small 3.1 large language model, Voxtral understands semantics deeply enough to support use cases like spoken Q&A, summaries, and even function calls triggered by voice. Voxtral is free to test via Hugging Face and available through Mistral’s Le Chat interface. For teams scaling audio applications, API pricing starts as low as $0.001 per minute, less than half of what major players charge for similar output. Mistral claims Voxtral Small matches or surpasses the performance of ElevenLabs Scribe and GPT‑4o‑mini, while Voxtral Mini Transcribe outpaces Whisper in cost-efficiency and speed. This launch follows closely behind Mistral’s “Magistral” reasoning models, further establishing the French startup’s push to rival closed-source giants like OpenAI, Meta, and Google. Voxtral isn’t just a technical achievement, it’s a strategic signal that open-source audio intelligence is now viable at scale.