Hi Synapticians!

Mistral AI has just made a splash with Voxtral, its new open‑source model that converts voice to text (and even to actions) at an unbeatable price. It comes in two sizes: a “pro” 24‑billion‑parameter version for servers and a lightweight 3‑billion‑parameter version that runs on a single computer. Voxtral is released under the Apache 2.0 license, so anyone can use, modify, and host it with no hidden fees.

What really sets it apart is the price: $0.001 per minute, about 6 cents an hour. At that rate, ten hours of transcription cost less than a cup of coffee. Proprietary services cannot keep up: OpenAI Whisper charges six times more, and ElevenLabs Scribe is roughly seven times pricier for the same duration. Yet Voxtral delivers comparable, if not superior, performance: state‑of‑the‑art multilingual recognition, automatic language detection, the ability to summarize or answer questions about the audio, and even direct action triggering (function calling) from voice, all with an extended 32 k‑token context that can handle up to half an hour of recording in one go.

Beyond the technical feat, the move is strategic. In a landscape where the EU’s AI Act demands transparency and control, Voxtral ticks every box: open code, on‑premises hosting to preserve confidentiality and cut latency, and above all a cost that democratizes a service long reserved for American tech giants. Mistral thus reasserts its vision of European “digital sovereignty”: with Voxtral, voice once again becomes the natural interface for applications, without sacrificing either budget or governance.

Here’s the rest of the news about AI today:

• Meta is contemplating a shift from open-source to closed AI models

• Google's NotebookLM has been enhanced with featured notebooks from The Economist, The Atlantic, and others

• Amazon has launched Kiro, an IDE powered by Claude, aimed at revolutionizing software development with AI

Top AI news

1. Mistral Unveils Voxtral: Open-Source Audio Model
Mistral has launched Voxtral, an open-source audio model that offers a cost-effective alternative to closed systems. It features accurate transcriptions and multilingual capabilities, with two variants tailored for different deployment needs. The API is available for free on Hugging Face, and integration costs start at $0.001 per minute. Read online 🕶️

2. Meta's Potential Shift from Open to Closed AI Models
Meta is contemplating a shift from open-source to closed AI models, which could significantly impact the AI landscape. This change might slow the momentum of open-source development and consolidate power among major players with closed ecosystems. The decision reflects Meta's need to monetize AI investments while maintaining competitive performance. Read online 🕶️

3. NotebookLM Expands with Notebooks from Top Publications
Google's NotebookLM has been enhanced with featured notebooks from The Economist, The Atlantic, and others, allowing users to explore a variety of topics with sourced answers. The platform now includes interactive features like Mind Maps and Audio Overviews, making it a comprehensive tool for research and learning. Read online 🕶️

4. Amazon's Kiro: A Claude-Powered IDE Revolution
Amazon has launched Kiro, an IDE powered by Claude, aimed at revolutionizing software development with AI. Kiro facilitates the creation of secure and scalable applications by combining project specifications, technical architecture, and automated task management. Now in public preview, it offers free monthly interactions and supports various platforms, making it a versatile tool for developers. Read online 🕶️

Meme of the Day

Stay Connected

Feel free to contact us with any feedback or suggestions; we’d love to hear from you !