What has happened with Gemini 2.5 Pro? Why has the performance of the model downgraded.
I previously wrote a lab report with 2.5 Pro 2 weeks ago. Yesterday I did one last check on the last report to find anything missing or wrong.
It was like a completely different model. It would consistently miss elements or typos it would normally wouldn’t miss. It’s almost as if it doesn’t even try to read my lab. It would just skim it. It would skim my detailed instructions. It would skim everything.
Hallucinations, laziness, lack of detail. I spent 4-5 hours trying to troubleshoot why it kept missing things.
I gave up and switched to claude 4. It was able to follow my instructions easily and I was able to fix the work in 30 minutes.
I’ve been a long time Gemini 2.5 Pro user and I long believed that google would lead the AI. But if they just secretly downgrade models and hope users won’t notice, it’s hard to trust the API anymore. If they want to change the model they should make an announcement or release a new one and have the old one stick around.