They don't actually, they used to but the latest mistrals have been lacking. People in our LLM community do like them for finetuning but then often use the older mistrals and only for entertainment. For anything remotely serious its gemma and qwen neither of which are european. It may correlate with the european AI act that forces companies to only tune on licensed data, less data is a worse model and its noticable at least on the fiction side of things.