Mistral’s New AI Audio Model “Voxtral” Is Open Source

French AI startup Mistral has just launched its latest model. This one’s audio-based, and open source as well.

Voxtral is Mistral’s very first family of audio models. It’s positioned as a B2B service, with its open source code giving developers more control over deployment than similar high-end closed models.

Here’s what to know about the Voxtral model, which currently comes in two variants — Voxtral Small and Voxtral Mini.

Voxtral Small vs. Voxtral Mini

According to Mistral, Voxtral Mini is the cheaper option, while Small is the premium version.

Both models include a range of impressive features including long-form context (32k token context length) and built-in Q&A functions, and they are natively multilingual. Spoken prompts can trigger actions in backend functions, workflows, or API calls.

The company’s official announcement sums it up:

“For cost-sensitive use-cases, Voxtral Mini Transcribe outperforms OpenAI Whisper for less than half the price. For premium use cases, Voxtral Small matches the performance of ElevenLabs Scribe, also for less than half the price.”

Mistral: The Scrappy Alternative to ChatGPT and Gemini?

This isn’t the first time that Mistral has debuted an AI model aimed at undercutting the biggest heavy hitters in the LLM industry. Back in November 2024, we covered Mistral’s new image generation and web search functions — both clearly positioned as ChatGPT rivals.

Now, the new Voxtral model is explicitly taking on Gemini 2.5 Flash. The official announcement even includes a chart that lines up the two Voxtral models alongside Gemini 2.5 Flash and two OpenAI tools, Whisper large-v3 and GPT-4o mini Transcribe.

According to Mistral’s analysis, Voxtral Mini delivers about the same word error rate as Gemini 2.5 Flash for a much lower cost, while Voxtral Small goes the other direction and lowers its error rate in comparison to the Gemini competitor, while costing a fair amount more.

The International Solution

The new models are multilingual — users can speak in English, Spanish, French, Portuguese, Hindi, German, Dutch, and Italian, among other languages — so they’re a fit for international businesses and audiences.

With the AI hype train still steaming ahead in 2025, Mistral is hoping to remain in the center of the pack as an attractive mid-range option.

Its Microsoft investments won’t hurt, either: The AI-hungry tech giant has given Mistral €15 million (or about $16 million) for a multi-year deal that would bring Mistral Large to its cloud computing platform Azure.

Voxtral pricing starts at $0.001 per minute for API calls, although users can download the free version now on Hugging Face, too.