I pay for Mistral even though I barely use it these days since I am not extremely fond of AI in general. It is however the only non local AI I will use whenever I use one, I absolutely refuse to give any of my thoughts to US or Chinese models.

I’m a paying customer and I use it the whole day. I’m pleased with the results, and happy to support EU tech. (BTW, don’t complain about Big Tech, and then refuse to boycott them because “they are simply better”).

Some of big European companies block usage of mistral ,but they do use USA models .

To be frank, I would want Mistral to be a huge success. But if open weights and an EU firm (sovereignty) matter to my project, I’ll just go straight to open source models like Soofi.

Does anybody else have a similar experience?

I like the concept of mistral and its good for most things. Coding on the other hand is difficult. Mistral medium 3.5 often fails to use its own tools to edit files, gets things wrong and ist rather slow. I just had the opportunity to test codex and I have to say its a lot stronger obviously. I hope mistral soon releases a new large model with better instructions flooding and coding skills and improved tool usage!

The only unfortunate thing is there’s a paper where the authors tested a variety of models against being prone to (for example Russian) propaganda. And Mistral models were especially prone to amplify Russian propaganda.