Mistral (family)

Mistral AI's model family — Mistral 7B, Mixtral 8x7B/8x22B (sparse MoE), Mistral Large, and Codestral — Europe's flagship LLM line, mixing open and commercial releases.

The Mistral family is the LLM line from Paris-based Mistral AI. Notable models: Mistral 7B (2023, widely regarded as the best 7B model when launched), Mixtral 8x7B and 8x22B (2024, popular sparse MoE), Mistral Large (commercial frontier), Codestral (coding-specialized), Mistral Small / Medium / Large (commercial tiers), and Pixtral (multimodal). It matters because Mistral models punched above their weight class. Mistral 7B beat several 13B competitors when released, demonstrating that careful training and architectural choices could outweigh raw parameter count. Mixtral 8x7B brought MoE architectures into the open-source mainstream — only ~12B parameters active per token despite 47B total, giving 13B-class speed with much better quality. The early models' Apache-2 licensing made them a default backbone for commercial applications. The family has evolved with mixed open / closed strategy: smaller models tend to be openly released, frontier-tier Mistral Large and similar are commercial. The company partners with Microsoft to host models on Azure, and runs the Le Chat consumer assistant. For European builders concerned about data sovereignty and US-China dependency, Mistral is a frequent choice. For self-hosting on consumer hardware, Mistral 7B and Mixtral remain popular default options. Related: Mistral AI, Mixture of Experts, Llama family, open-source.