AI quality dropping recently

Google AI Studio
,
1

Hey guys. For the past 10 days… it’s really bad.
The AI ​​has been acting like a 3-year-old.
Gemini 2.5 and Gemini 2.5 Pro. And I noticed that it’s losing focus before the token count even reaches 300,000… The AI ​​was really great until 10 days ago. Now it’s really bad.

2 Likes
3

Hi @python
I apologize for the frustrating experience.
Can you provide some specific examples or types of tasks where you’re observing this? Is this happening in API or AI Studio?
Thank you

5

"Hello, :waving_hand:
First of all, thank you for responding to me.
I have been writing AI-supported Python code in Colab for approximately 8 months. I used to solve the problems I couldn’t handle in my Python code using Google AI Studio.

BEFORE -)

  • I could upload the entire code to the AI and get solutions when I asked questions.

  • It was as if the AI internalized my code.

  • The AI could understand the question I asked and even confirmed it by reiterating what it understood.

  • I could get CLEAR and EXPLICIT solutions to the questions and problems I asked the AI.

  • If the AI occasionally lost focus, it would regain it.

  • I was getting good performance within the limited 1 million token limit, around 650-700 thousand tokens.

  • We were making arrangements within the agreed plan regarding the code, and it adhered to the plan.

  • The AI was smarter; in terms of answers and reciprocal Q&A, it was as if there was a truly advanced intelligence in front of me.

  • It followed the instructions I wrote in the ‘instructions’ section.

  • In terms of conversational context, it remembered longer (previous) questions and answers.

NOW -)

  • I want to say this from the outset… now I know that if I ask the AI something, I won’t get relevant or correct answers.

  • The connection completely breaks after the token level reaches 260 thousand.

  • It gives irrelevant answers.

  • It contains very long answers and details.

  • The answers it provides contextually remember at most 2 previous conversations.

  • It constantly says it’s providing incorrect information and apologizes for it.

  • The conversation it initially understood correctly later leads to completely different answers, and we go off-topic.

  • The AI understands the topic correctly; I make it repeat what it understood… but it deviates in application.

  • It does not follow the instructions.

  • The most important issue is that it cannot establish CONTEXT.

I apologize, these are not complaints; I am someone who trusts and likes Google products. Perhaps I am expecting too much, but I think there might have been some issues after the last update. I wish success to the Google engineers. Regards. (Country: Turkey)"

7

This is it! I tested Gemini 2.5 Pro today…
The engineers at Google have gone so far this time. It’s as if the AI is reading my mind… it understands what I want, it doesn’t constantly make mistakes and apologize, and it doesn’t write excessively long texts. The most important thing is its ability to establish context and not lose focus on the topic.
It’s genuinely fantastic.
GOOGLE AI STUDIO… is perfect. It even offers me suggestions for the problems I’m trying to solve. There was a bug I had been working on for months. Today, Gemini 2.5 Pro solved it.
Thank you, Google. Thank you to the Google employees.

8

You just need to wait a little bit because it will find a way to screw things up again. It’s worse than dealing with your spouse sometimes. You never know what you’re going to get, depending on what side of the bed it wakes up on.

1 Like