Cursor with Gemini 2.5 Pro (exp AND max) used to be beast until April 22nd. I don’t know what happened, but suddenly Gemini started to degrade in performance to an unusable state.
At the core of the issue seems to be that “thoughts” often - though not always - no longer get captured in the thinking bubble; instead, they appear as normal output which gets reprocessed again and again, until the model falls apart, often repeating itself infinitely.
I can’t tell whether this is Gemini’s fault or Cursor’s, but it has rendered Gemini useless in Cursor.
2 Likes
It looks MUCH better now, will keep an eye on, obviously. Wasted about 100 USD on this… inspirational output.
Are you finding it worse after the outage? It is working but it is way dumber, forgetful, makes the same mistakes, not detailed, missing things. How unfortunate 
Curious if you’re finding the same?
Yes, indeed! I find myself using Sonnet when it‘s even a bit more complex. I often interrupt Gem‘s poking around with a simple: ‚Take over, my dev is stuck.‘ In 90% of cases Claude just gets the job finished.
Gem is super fast, but has become unreliable,
It sometimes or often misses the most obvious things.
Yes, it’s definitely still useless after the outage. I got no credits back from the cursor team.
It seems like the team don’t care about this and just want you to burn through requests.