Hello! Recently, I stumbled upon a really interesting behavior of the model.
I uploaded a YouTube video and gave Gemini 2.5 Pro a simple task - to analyze it.
This is the model’s output, nothing in the chat was edited or altered in any way:
It was thinking, then generated one part of the response, then paused to think AGAIN, then generated the next part, and so on.
I find this behavior very promising and just wanted to let the devs know what their model might be capable of.
Hi @klinok64,
The image is not clear and pixellated.
Can you re-upload the image or take multiple normal sized screenshots and share it.
Of course. Here it is, non-pixelated
Hello
Welcome to the Forum!!
Thanks for sending over that feedback and the screenshot regarding Gemini 2.5 Pro. We tried to reproduce this on our end, but the model processed the video and gave us a summary immediately, without showing the multi-step ‘thinking’ process you mentioned.
Could you let us know which parameters or configurations were used during the video analysis?
