#Gemini 3 pro feedback

3 messages · Page 1 of 1 (latest)

fading pollen
#

I'm not sure if I am supposed to give gemini 3 model feedback in this server (as it's mainly focused on gemini app), but here are some of my thoughts:

The only three things I don't like about gemini 3 is:

  • Instructions following
  • Token Limit (cant write above 800 lines code, and tends to write 400 lines code)
  • it tends to write the slop UIs (similar to 2.5 pro) unless the prompt is directing a different style.

Otherwise, gemini 3 is my favorite model.

Would really appreciate if this could be passed along to the team!

grim kettle
#

My Main Concerns
The main issues I have are similar (across different interactions). Gemini 3 tries to find a solution too quickly. As soon as it hits its token limit, it repeats output without changing anything. It cannot handle old chats and mixes up memories.

It reacts very stubbornly; if it has once fabricated nonsense, it sticks to the wrong opinion for the rest of the chat and repeats the error.

If you request a large quantity of output following a certain structure, the output generation stops, and it renders the same canvas over and over again.

It generates images and videos without being asked, even when they are not desired.

Overall, it is very clever—that cannot be denied—but it feels like it only has a short-term memory. It can handle a single request well, but as soon as things get more complicated, it seems hopelessly inferior to the previous model. For my purposes, I would prefer to have version 2.5 back until this is resolved.

dawn latch