Do they? That's surprising. I saw them come up here twice for their OCR model, I tried using it on a 200 PDF that was just printed text without embedded OCR and it failed miserably - got less than tesseract and I ended up with a $5 bill.
I figured Mistral was a nice idea and liked because it was a European competitor more than because it competes. I'll be happy to be wrong if it has improved
I tried the API on a coding task and it did worse than I would've hoped. Also $5. They still have a ways to go, I think.
I also use their chat pretty frequently, it seems to "misunderstand" me more often than other models. But that's free and it isn't as irritating to rephrase a question as it is backing out code changes.
I like that they have local models available, I plan to try those sometime.