Increase the limits
What feature would you like to see?
I've been using codex for the last week.
I'm paying a teams subscription and my friend, who doesn't code and has a plus subscription has allowed me to use his, because I reached my weekly limit.
So that's 30€ + 23€ = 53€ a month worth of subscription, for reference I'm running a small 1 control plane 3 worker k8s cluster for the same price.
If I had a steady influx of money I wouldn't mind, I'd even go Pro.
It seemed like the plus subscription lasted longer than the more expensive teams subscription.
Both lasted about 2 full days each, which is not enough. And 53€ / month is not exactly cheap.
I can't afford 230€ / month for Pro, that's half of what I'm paying for my apartment, and I'm living cheaply compared.
Limits being set this low, what's the point of subscribing at all, when you can't even use it a full 5 day 8 hours a day work week? But in reality I'd use it for 12h 7 days a week, especially when in the middle of something.
The problem is also that the LLM doesn't get it right quickly, and you have to tediously go back and forth.
Example a migration from an external server to using ko build with argocd yesterday took about 12 hours, hit a stopper, refine, goto 10. 5 hours for the first one, another 3 for the 2nd one, 2 hours of random code improvements.
And you constantly have to remind it to either use code blocks, because the formatting is off, or to not output code in the chat and do it on its own, or to not make the same mistakes again and again, all burning tokens. And then it provides option A or option B, which is far too complicated. And it introduces a complex solution and tries to work on that instead of keeping it simple.
I even suspect this is by design, to keep customers burning tokens.
At least it's not making "typos" like regular ChatGPT. That one simulates a human by "making typos". "Oops that was a typo". What the hell?
And codex has no support for multiple accounts, so I have to manually switch the auth.json around.
The limits are set too low and need to be increased at least 3 fold.
Are you interested in implementing this feature?
Good one
Additional information
_No response_
3 Comments
I am also facing the same issue. The solution would be for the project to implement a model routing feature.
Since 4o-mini can be used virtually without limits, I want simple requests to automatically switch and use the mini model.
Cursor's AUTO mode is already doing this, so it should be possible, right? For example, you could just route requests to the virtually unlimited mini model and let it decide whether to use GPT-5 or handle the request with mini.
If this feature is implemented, it would greatly help many Plus users, so I hope it will be prioritized.
Believe it or not, but on the contrary, we are working daily to figure out how to pack more intelligence into fewer tokens and to have the model use the minimal amount of thinking necessary to get the job done.
Going to close this as we are actively working on a few things here to allow getting more usage out of codex. Very excited to keep pushing on how much work you can get done within the bounds of your subscription.
On the "manually switch the auth.json around" part:
aiswautomates exactly that. Named profiles with isolatedCODEX_HOMEdirs, one command to switch, backs up credentials before every switch. Also works for Claude Code and Gemini CLI.https://github.com/burakdede/aisw