Hi everyone, new to Dev forums. Please forgive me if I am posting in the wrong Forum and if so, kindly direct me where to correctly post. We are enormously frustrated with Google Gemini’s Ultra promises. I detail our issues and frustrations below. My hope is Google’s Gemini app team looks at these posts. We have lost a week, wasted Gemini Ultra-level app subscriptions, will suffer through alternate solutions without any refunds, and call it a day. But at the very least, our hope is the Gemini team understands frustrations from users that wholly wanted it to work. Here we go:
More Gemini Ultra App Inconsistencies. This time with “Thinking” Mode.
I think the folks at Google need to take a step back and decide if what they sell for the Ultra Subscription is something they actually are willing to allow.
Example: The Ultra app allows 1,500 “Thinking” level prompts a day. There are 1,440 minutes in a day. A human needs to sleep or not be in front of the computer for at least 4 of those hours. So let’s say you have someone with a crazy focus working 20 hours a day. It takes about 2 minutes per “Thinking” prompt to paste it in, have it process, then retrieve it, save it in a repository for review later, etc.
Assuming that in a 1,440-minute day you have 20 hours you are working, that is 1,200 minutes you have available. Given that it takes at least 2 minutes per prompt from start to finish to be initiated, processed, and thereafter logged in some fashion, you are looking at 600 prompts max that can be executed. So that leaves you to opening up multiple browsers and cranking through them if you don’t want to be up for 20 hours. (I think automating the movements is against the rules.)
The kind of people that sign up for Ultra aren’t casual users. They usually have very intense and focused objectives. So they don’t move at happenstance speed. Yet the app, from our experience, seems to be designed for the casual “Pro” happenstance speed.
We’ve been trying to crank out “Thinking”-level prompts. At about every 100 or so we get throttled, forced to wait from half an hour to an hour, log out, log back in, repaste the prompts.
So the question is: does Google want to charge Ultra-level prices and advertise super-human production capabilities, but when users who actually push those advertised limits take the offer to task, not deliver? Is the available capacity there for show, or does Google actually want to deliver on the promised capacity they make available?
I don’t think there is anything nefarious going on here. I just wonder if the folks at Google in charge of the Gemini app have actually executed the use cases they charge for, and advertise as, available. Because if they did, they would get throttled.
Lastly, we’ve been frustrated with the Gemini app and the disappointing Gemini Ultra responsiveness, so we attempted to revert to the API. But the results in the API aren’t as sharp as “Thinking” inside the App. They are actually 50% of the quality of the outputs in the “Thinking” app. We were quite stunned when we took outputs from the same prompts and compared.
So we had to scrap the API option for our cases. Why the Google Grounding and other APIs for basic search are worse than the search queries initiated in the app, for us and our use cases at least, is a mystery.
We recognize things are moving fast in the AI space. That enhancements, corrections, improvements, are made hourly to daily. We are enormously frustrated and my hope is the folks at Google read these forums. We welcome speaking directly to anyone at Google about these things and we recognize these things may already be on their list of things to examine.