By skyforbes - Nano Banana Pro – good for one shot images – awful for editing or changing aspects of those images. There seems little point using editing prompts. Just have to start over again. Use the image you liked as a reference and hope the changes are made.
-
Veo 3.1 – unbelievably dumb at times. Basic errors where for example you tell it only one character in the scene talks and end up with both characters sharing the dialogue.
-
Google Deep Think – Gave it a complicated research task and after 20 minutes it returned with nothing more than a 700 word chat reply. Incredible frustrating.
-
Google CLI – no way to get extra usage if an Ultra subscriber. You pay 250 dollars a month and can't use that for coding via the CLI?
-
Confusing eco-system. Should I use the app or AIStudio, NotebookLLM, Flow, API, Vertex?
-
Incredibly poor instruction following – There are times when it just doesn't get the semantics – there's something wrong with its inference parameters – I'm talking Gemini pro 3. You have to be super specific immediately. It seems to lose track.
-
The Ultra plan. There is no value at all in the plan. Even for video and image creation which is hit and miss and you can waste hours just getting a short project to work out.
-
Gemini 3 – it's a good model but it's not easy to use it in a way that you see so much difference between it and Gemini 2.5. I'm not getting that much better output from Gemini 3.
In all I find Google incredibly frustrating. They have a scatter gun approach – no unified vision. No singular point of focus. My suggestions are that they unify their offerings in one portal. Offer smart tool usage. Work on inference. Have their models follow instructions. Allow Ultra users to have extra token usage in the CLI.