Gemini 2.0 Flash very erratic behavior

I am using Geimini 2.0 Flash using Vertex AI and everything was going great over 3 months or so. Suddenly, it is almost impossible to use now. Not sure a coincidence or not, but I am developing a code and my thread was filling up. So started a new thread and everything dropped off the cliff. Its like something drastically changed. The agent doesn’t remember anything, makes too many mistakes. Anybody experience similar behavior?

1 Like

Hey @Subodh_Nimkar , welcome to the forum

Just to clarify, it is the model quality that dropped, or is it more about memory/context not working right?

I am trying to confirm whether the results quality has dropped or not, but, my initial assessment is, it has. What I am certain about is the coding quality through chat has dropped tremendously. It doesn’t really remember even the last comment in the thread. Also, it used review code very efficiently and accurately, but now, it asks for very small snippets of the code. The reason gives, “my process doesn’t allow me to look at larger code chunks” or something to that effect.

To add, the latency (I think that is the right phrase or lag) has gone up tremendously. The responses in the thread are for prompts sent a while ago. As a novice, I was really enjoying the ride but now speed of code development has really dropped.

am trying to confirm whether the results quality has dropped or not, but, my initial assessment is, it has. What I am certain about is the coding quality through chat has dropped tremendously. It doesn’t really remember even the last comment in the thread. Also, it used review code very efficiently and accurately, but now, it asks for very small snippets of the code. The reason gives, “my process doesn’t allow me to look at larger code chunks” or something to that effect.

1 Like

HI
I can confirm that the model quality has not changed.
Subodh