How effective are Mistral's models at this point?
My perception using them is that they have comparable models to OpenAI and others when it comes to general use 'chat' tasks, but they don't match something like gpt-5 Pro or high thinking when you need something more powerful.
Perhaps the problem is that you need to be competing right at the frontier, or not at all. I put Cohere in a similar bucket.
I don’t know why you’re downvoted, this is an excellent question.
A year ago I wanted to use and like Mistral, mostly because it was an upstart competitor to OpenAI. Yet I found its coding ability sadly lacking. I haven’t tried since. I also have seen them on the HN front page very rarely. I’m curious too how well they stand up these days.
After testing it for myself, I find it's pretty solid for a lot of general use cases but basically breaks as soon as it needs to make multiple tool calls.
A bit of a shame if they're going to launch these connectors that the model isn't suitable for them.