Very dumb today, and also notice some reduction in thinking time.
Strugle for simple code implementation, had to go back with GPT 4.1.
It’s so annying, not to know what they are doing behind the scenes.
Until I can explicitely select this model, I won’t be using it.
I’ve noticed that many times today he’s failed the tool call parameter
Yeah, the tools call is so bad with that model, it fails most of the time, hopefully it gets fixed soon.
It’s nothing new cause 03-25 did the same, but I’m so tired of that “you’re absolutely right” phrase.
This is programming. AI needs to be able to do some critical thinking. Not to have “the user is absolutely right” as a default mode. And project rules are only partly helpful with getting rid of that.
And “soon” is an understatement. “Soon” would be already yesterday. All the tool usage issues with this model update, and how often it stops in the middle of a response, make Gemini very hard to use.
And wasting both users (fast requests) and Cursor (whatever their deal with Ai providers is) resources, because just getting through a simple prompt often requires 3 requests to work.
The gemini model is very fond of writing useless annotations 
after recovery from the massive degradation downtime today, Gemini 2.5 pro seems to be back how I remember it to be (ie. awesome and not always thinking)
Anyone find the same?