Gemini 2.5 is producing garbage as compared to 2025-04-10

Feedback
1

I’ve been using cursor for at least a year.
I usually don’t write,
In fact I never write, but…
this is bad..
I dunno what happened. but the output is REALLY bad as opposed to a few days ago.
Somebody did something, please revert…

How are you augmenting our instructions?

I know I haven’t done anything to my rules or instructions, I always try to be consistent, but the output I currently get is nothing like it was 24 hours ago:

  1. Tools are not found,
  2. Garbled code produced with no descriptions but just a partial snippet.

I now have to prompt twice saying “Try again” or “Fix it”, which doubles the already high enough cost.

The cost:
I’ve user 0 out 500 premium calls, all my calls are Max.. so what is the tier I’m paying a premium for? Also questionable.

This is bad. I don’t mind the 20 extra bucks, but not being productive for half a day is a Not an Option!

Fix this. I need to trust this tool.

1 Like
4

It is still the case Gemini 2.5 Pro Max has been super stupid since I wrote the post.

Agentic behavior is down and does not run for long, tools rarely run at all, Yolo is broken.

Before this, I would get messages saying 25 tool calls is the max, but not anymore.

Something happened, 100%.

@Cursor, respond please.