My Second Try with Mistral
Once again, it is time to reconsider which AIs I want to pay for. So I canceled my Google subscription — Gemini 2.5 Pro is indeed very good, especially when it comes to generating code, but the CLI doesn’t convince me; it’s simply not my way of working. And the other Gemini features, like its ability to generate endlessly long, pretentious 20-page analysis documents for any random question, are more amusing than useful.
And then, Google is Google, the data octopus. I assume that sooner or later all my chats will become part of my profile, which Google will use to deliver customized ads to me on behalf of advertisers. And Google is American. So, away with it — and once more, let’s try Mistral, the only noteworthy European LLM.
My last attempt with Mistral was about a year ago and rather sobering. Getting answers that are as correct and as free from hallucinations as possible is essential; an AI that is significantly worse at this than another is useless. Now Mistral has Memories, Projects, Reflection, and Research — so it’s time to give it another try.
The first task: write this blog entry. I provide the plain text, and it transforms it, based on the CSS and another HTML entry, into HTML, both a _de and an _en version. And as a reference, I will do the same with ChatGPT 5 and with Mistral.