Personally, I don’t see any issues with the limits:
-
It resets every 5 hours.
-
My code is very modular and granular.*
-
I provide context for the AI to work with.*
-
Claude models are not their own models, so I use them once in a while and, while I would want higher limits, it’s understandable why they are so low.
-
The 5-hour reset is really good because it renews while you are working.
-
I don’t let it go free and search for solutions available online, I limit it by using specifics like: Library, Framework , APIs, SDKs
-
There was one time when I had to wait an hour for the limits to reset, and I used 30 minutes of that time to take a break and do things I enjoy. Then I came back and made a summary of everything I had done.
I personally** take it as: if my limits expire while I’m working, it means that I am doing something wrong or that I should take a step back and review what I’m doing.
* This avoids wasting tokens searching for stuff.
** My POV only - not implying anything beyond that.
- Did you create an context file for your model to read and know what to look for?
- Did you limit what it can use/do ?
- Did you prepare the model for the specific project?
I’m not trying to lecture anyone or criticize the way others use it. I’m just genuinely surprised that so many people are struggling with the limits. I’m also open to advice and tips from others.
The reset is every 5 hours, not 4 hours.
On simple projects I never hit the quota. On complex project I do reach the quota limit quite often if not always, Is that simple. I only need agents for complex work. If you aren’t hitting the quota limit then good, you know the answer.
Edit: It seems today they added a low reasoning/thinking profile to Gemini 3.5 Flash, let’s see how it behaves.
Same, I upgraded to AI Ultra and it’s been breezy. First I got a notice on Pro (your weekly quotas have been reset. Also your quota limits have been increased by x3), was nice, but I needed more, so I upgraded (specifically for the free $100 credits with ultra), I am struggling to saturate a 5h window…Across 6 projects running at once, across 3 systems…
I’m not happy today with the limits.
Just wait till you get hit with a multi day limit refresh when you hadn’t hit a single 5 hour limit in 4 days and usually had 40% or more unused.
And this is just having the AI help add options to streamlit or minor things like helping write code to have Python program go through a data file and calculate averages of certain data points.
Absolutely nothing that would be resource intensive for the AI.
I am also happy with the new limits. An yes, I have all my harness sanitazied in many ways. Two weeks ago I was not even reaching the 5 hours limit before my weekly quota was gone. This is fixed now for Gemini models. As for Claude, I can barelly use Opus for a couple of prompts and the weekly was gone. I dont mind, use Claude only for helping in Skills, MCP and oprimizations of the harness itself.
It was inevitable that pricing changes were going to occur. Personally, i dont mind it. and if i go over, i pay for it. 
Right now, I am not the biggest fan of the limits. I am doing my best to reduce token consumption, and hit the limit pretty quickly. I cannot afford a higher tier plan or more credits, so I just do something else for a few hours.
I tend to get 90 minutes to 120 minutes of work done before I hit the limit. Yes, it is a growing project. Yes I make sure to start new sessions to keep context lean and easy. I am fairly descriptive, and the model does what I ask with little to no refining.
