LOADING...

Mistral drops 10 new open-weight AI models

Technology

French startup Mistral just released the Mistral 3 family—10 new open-weight AI models.
The lineup features a powerful flagship model and nine smaller, customizable ones built for businesses and edge devices.
These models are designed to run efficiently, work offline, and can be fine-tuned for different needs.

What's special about Mistral Large 3?

The top model, Mistral Large 3, packs 41 billion parameters and handles huge context windows (up to 256k tokens). It supports both text and images in multiple languages thanks to its Mixture of Experts setup.
The smaller models (with 14B, 8B, or 3B parameters) can even run on a single GPU—perfect if you want AI that does more than just chat.

Why should you care?

Mistral is all about giving users control without forcing them onto pricey cloud services—great if you want privacy or flexibility.
The focus is on cost efficiency and easy customization.

How does it stack up?

Mistral Large 3 goes head-to-head with big names like OpenAI's GPT-4o and Google's Gemini.
The rest of the lineup competes with Cohere Command A and Meta's Llama 3—but stands out by letting you run everything offline with open weights instead of relying on the cloud.