Mistral 3 – Open-source multimodal models with scalable performance

Summary: Mistral 3 offers three dense models (14B, 8B, 3B) and a sparse mixture-of-experts model with 41B active parameters, all under Apache 2.0 license. It delivers high performance-to-cost efficiency and includes instruction-fine-tuned capabilities.

What it does

Mistral 3 provides state-of-the-art dense and sparse models designed for multimodal tasks, featuring a large mixture-of-experts model trained with 675B total parameters.

Who it's for

It targets developers and researchers needing open-source, high-performance multimodal models with scalable parameter sizes.

Why it matters

It addresses the need for cost-efficient, powerful open-source models with instruction fine-tuning for advanced applications.