I’m on the $200 Pro plan, and my weekly limit was completely exhausted in 2 days of serious development work.
This is simply not viable. When someone pays for the highest consumer tier, the expectation is capacity for sustained, professional work — not being locked out for the rest of the week after two days.
Being blocked for hours already breaks a workflow; being blocked for days makes the tool unreliable for anyone doing real development rather than occasional queries.
What makes it worse is the framing. The plan is sold with language suggesting far more headroom than what’s actually delivered. In practice, the gap between what’s advertised for the higher tiers and what the weekly cap allows is hard to reconcile — and I know from the community I’m not the only Pro user hitting this wall far sooner than the price would suggest.
My questions to the team are direct:
Will the usage limits for higher tiers (Pro 5x/20x) be reviewed? The current cap makes longer, serious coding sessions impossible to rely on.
Can we get transparency on how the weekly cap is actually calculated, given that usage counters seem to move disproportionately to the real work done?
I want to keep using Codex — that’s why I’m on Pro.
But at these limits, the higher tiers are not sustainable for the exact professional workflows they’re supposed to serve. I’d genuinely appreciate an official response on whether this is being looked at.
Same… I went to bed last night and had around 70% remaining after a Saturday reset, came back this morning to see 0% remaining. Not like I had something running overnight, and yet all usage blocked until this next Saturday. Hard to trust a $200 / month service that is completely unpredictable about whether you’ll even be allowed to use it for most of the week and zero recourse for “is this an outage, or did the calculation just change? In any case, figure out something else to do for the next week.”
Welcome to the developer community, @jjjnoronha and @kalani, and thank you for taking the time to share such detailed feedback.
It may be helpful to review your workflow, particularly how frequently subagents are created and how tasks are delegated, as these patterns can affect overall usage.
Where appropriate, consider using smaller models, such as Terra or Luna, for routine or less complex tasks, while reserving larger models for work that requires deeper reasoning. This approach may help extend your available capacity without significantly affecting productivity.
I’m a professional developer on Pro and I worked very hard over the weekend on about a dozen PRs using Sol High. I depleted my account over 3 days by about 12%. So I agree with SPS, something is off with your workflow. Are you playing fire and forget one shots and then reviewing at the very end and dealing with a lot of rework or are you steering the model as you go along?
Are you spawning lots of subagents with significant autonomy? e.g. on Ultra? If so I’m not surprised if that might burn through a lot.
Just wait til there is no subsidy. $200 is currently a bargain - it is rumoured it costs OpenAI $5000-$10000 to provide you that service.
I have the same issue im so pissed i dont know what do the main issue is the weekly rest keep being pushed 2-3 days further so today i ma left with no usage even though i have input more than 300 $ on top pf my pro max subcription
Nope I was doing hard work but only on 1 project I had 2 agents running one coding plus another checking the code quality, I am refactoring a Rust backend.
I had a 3 agent but is on claudecode he checks a .MD file and scans for more critical issues.
I did forget the ultra mode about 3,4 hours of code that alone flew 30% of weekly usage.
The rest I was using Sol -Medium.
I am not an expert in codex or in any IDE I am migrating from QA tester and Tests on Opencode to a full Ai Engineer with system Design software engineer etc.
ATM I am building something real for a custumor so I am a bit cautious and burning more tokens on security things and doble and triple checks.
The fact that I was working on a solo project was what made me feel a bit sad, because I can’t sustain 600d month, and shifting for Opencode is unviable due to bad multi agent review.
I will probably try claude code 200d and Olama Kimi 100dx2 if needed.
I can’t sustain more then 300dolar a month I’m a freelancer no resources I’m starting.
Yeah, I would agree, for freelancers surely $300 is a practical limit for most? I think OpenAI marketing team would understand that, hence the current pricing. OpenAI have time, but they are going to need to increase efficiency to a point where increasing the sub won’t be necessary because they can bring the token cost and token counts down for the same level of service.
I opened a second $200 / month account Yesterday and went to bed with 91% usage remaining. NOTHING ran overnight, and my laptop was asleep and has NO scheduled tasks configured within ChatGPT / Codex.. This new account was only signed in on my laptop and nowhere else, and only via the ChatGPT Desktop app on macOS.
This morning, I woke up and this account too shows 0% Usage remaining until Monday, August 10 at 11:40pm. So, paid an extra $200 for a second account to get 9% usage for the week, and now BOTH accounts have 0% remaining.
And, of course there are zero metrics or data that I can use to check anything… I just need to take it on faith that things work and are accounted for properly despite now 2 accounts having the same issue.
In the codex app, you can look into your profile to see the daily consumption of tokens in the grid chart. Hover the mouse over the days to see the numbers.
That might give a better hint on the level of usage for the tasks in the day the usage was drained.
Sometimes “1 task” or “1 repo” doesn’t mean it is a simple job, and the usage can be very different for each project.
yeah i checked now it seems i spent 3bi tokens in 2 days.
im not sure how i alredy have embeding and stuf to help on token wast.
There is a possibility that the 2 agents constatly looking to all backend because its a refactoring they burned a lot of tokens i dont know.
if thats the case maybe I would need a open source to code or maybe Grok4.5 or and have gpt only wacthing.
whats stresses me more i not being able to predict or make projetions, i could go to the custumer and say i need more x or y. But i cant because its not linear.
yeah, I get it. Hmmm … I think you need to work on optimising that workflow so you can stay more sane . That said, difficult when the tools and token usage fluctuates.
Omg finally I have someone that has the same issue like omg this is such a situation that bring anxiety I have multiple clients deliverables in progress ( it’s almost like the new instructions were burn as much usage token but don’t deliver nothing
You know you’re in a bad spot when a $30 grok subscription looks generous by comparison and letting me do considerably more than I can do with 2x OpenAI Pro $200 subscriptions.
Something is broken about usage limits, and in a bad way.
You all should be using Terra only for 90% of tasks, Terra is 40% the cost of Sol, Luna is 4% the cost of Sol. It’s not difficult to optimize your usage, especially if you have multiple agents working at the same time. Not only that Sol takes a lot more time to complete a task, especially on Ultra. You’ll use much less time and tokens on Terra.
Unfortunately this is just the way it is. We have no understanding of what the limits are outside of a percentage.
One could argue that the current limits are more than fair as their API-usage, I would argue that it’s encouraging unrealistic usage in such a way that encourages people to continuously upgrade into a tier that eventually will have the same performance as their previous one.
A 20x Pro Plan running out in 2 days is still a serious issue.
I did a full day sprint with Codex on the regular pro plan and was only knocked down roughly 15%. No sub-agents or automated processes. The harness and the model really do a fantastic job of untangling the chaos of codebases.
Same… I went to bed last night and had around 70% remaining after a Saturday reset, came back this morning to see 0% remaining
This is crazy. I wonder if there’s some loop going on. I have noticed that the model tends to implement “production-grade” code. Error consumption, code hardening. Just needs a single “Wait, let’s do this instead” to now have numerous dead modules that another model can accidentally fall upon and update.
Even worse, these dang models cannot stop with redundancy in testing. I can’t tell you the number of times I’ve seen the model ask to run headless chromium browsers. Not happening
I had 39 Sub Agents Active on my project, they all were doing nothing but timer was counting 20+ Hours
Tried multiple times to delete them with help of model, a lot of alucinations, and i did not understood clearly the documentation, looked buggy the subagents stuff
managed to stop them
Downgraded the model to Terra Medium
Run Program
100% exhausted in 9 Hours
Downgrading model did 0 to Rates.
Then i went to the dealer and grab all co** Possible.
I felt much better so i decided to threw more 100d so 2x Pro Sub.
Shut Down computer for several hours
Felt much better with or without Substances
Downgraded model to 5.5(medium)
restarted workload
10 hours only 4% exhausted in Fast Mode
Tried switching back to Sol (Medium)
Runned for more 7 hours ( only 3% more)
(After Sub and Restart) 17 Hours Work 7% Vs (Before) 20h Exhausted
switching models don’t seem to have diference on rates. (weird)
Subagents Fix only worked with shut down of computer not restarting codex ( and i mean shut down on console and checked on task manager)
My Opinion: As a Tester i have that Feeling when something is not right and i have this feeling here, something on the subagents don’t look ok, i cannot test because it would mean several $ spent in Testing, because was kind of what i was doing.
I stil learning all this plus working 16h+ a day trying to grab all this Ai World.
a lot of things i don’t have confidence to say “this is a bug”
EVIDENCE OF SYSTEM BUG: 26 auto-reloads in 44 hours totaling $466.82 on a $200 Pro Account.
My billing profile just suffered a massive auto-reload loop. Between Wednesday afternoon and this morning, OpenAI charged my card 26 separate times (Invoices 029KQIEP-0005 through 029KQIEP-0029) for minor script updates using the Codex/Spark interface.
At peak malfunction, I was billed at 4:10 PM, 4:32 PM, and 4:50 PM consecutively. The system is burning through credit reloads in 20-minute windows while the front-end UI hangs on “thinking” loops without generating outputs.
This is proof of a critical backend token leak or an unchecked context replication issue with the new agent structures. I need an engineer to audit these specific invoice sequences immediately.
If Sol Ultra is already having a hard time with some of the tasks I give it, Terra simply isn’t a realistic replacement for 90% of my workload.
I’m working on a real production system with 250k+ lines of code across roughly 300 files, with interconnected business logic, database rules, permissions, integrations and dependencies.
For simple tasks, isolated functions, POCs or repetitive work? Sure, Terra makes sense and optimizing model cost is smart.
But on complex changes, the cheapest model isn’t necessarily the cheapest solution. If I need 3–4 attempts, more supervision, more debugging and then Sol to fix what Terra couldn’t understand, I didn’t save anything.
For me the optimization is not cost per token. It’s cost per correctly completed task.
And on a large existing codebase, context, reasoning and architectural understanding matter a lot more than raw token price.
. That said, difficult when the tools and token usage fluctuates.
I have multiple clients deliverables in progress ( it’s almost like the new instructions were burn as much usage token but don’t deliver nothing 
