In the past, I used to be able to use about 2-3 billion tokens per week without hitting a usage limit. Yet, for about the past month, I have been /lucky/ if I can get 488 million tokens per week before hitting a usage limit, and I have had some weeks where I get told I’ve used it all after 38 million tokens. Based on my usage and data available under Profile, I objectively seem to have approximately 1/10th the usage I had this time last month, and am regularly hitting limits within 2 days of what I’d consider fairly light work with no sub-agents and on 5.6 Sol High. Unfortunately, no one at OpenAI seems to be saying anything about this. Opening support tickets, and I am told that I should wait until my next usage reset, consider reducing my usage, or moving to a lower tier model like Terra or Luna. When I ask expressly why my usage limits seem to be 1/10th what they’ve been for years now despite continuing to be on their $200 / month plan, I receive no response.
Something is off.
Either your domain is a relatively expensive one to work in or something in your workflow or architecture isn’t optimal.
Is your project open source?
I’m not claiming to be any kind of guru, but even going really hard I cannot destroy the quota in half a week let alone two days.
Hey, I’m a Plus user, and before July 30th I could use about 150M tokens on a weekly basis. After that date, the maximum tokens I’ve used before fully draining my weekly limit is about 50M. So there is a bug or some limit changes by OpenAI.
When there are threads with 431 messages in them describing this identical behavior across 150 users … maybe it’s actually an issue and not just 1 person’s anecdote about a less optimal workspace or architecture? Codex Rate Limits Discussion Thread - #448 by kalani
I feel the same. I’ve the $200 plan and my weekly limit was reached in 2 days
I’d like to see more evidence.
What’s the tech stack, what were the prompts, what were the code changes, etc.
Share your detailed workflow, share your repos!
Because I’m simply not experiencing this issue and I’m fascinated to see what would cause it.
No, this is not a bug. It happened to me. I use 5.6 Sol with High/Very High only, not even been 2 days already finished. This has been happening since 9 August, the first reset of my usage. Then, 2 days later, my usage finished again, and I have to wait until 20 August. I was a Claude Code user who moved to Codex, hoping for a new fresh start, but then I got slapped with Corporate reality like this. All they want is to drain our money and empty our pockets. What a waste.
PS: I’m a 5x $100 user who uses Codex for creating motion graphics and animation from code.
Is anyone at OpenAI looking into this suddenly insane token burn-rate? I had two refreshes banked and they vanished. Purchased credits seem to be being consumed at about 1000 per 30 minutes on 5.6 Sol Extra High.
Not open source, it’s proprietary multi-layered data analytics with some from-scratch bridging in between existing analytic software. Token usage is painful.
The project isn’t public—it’s internal company work—so I can’t share the repository. It’s also not some massive codebase with thousands of files.
I actually used to be one of the people asking, “How are you even hitting the limit?” With the same kind of workload—simple fixes, UI adjustments, and small tasks—around 100 tasks would previously take my usage from roughly 100% to 90%. Now, even 50 similar tasks can bring it down to around 75%.
And before assuming this is because of heavy prompts or because I constantly use Ultra, here’s the important context: I’ve been a Pro user for 5+ months and have worked this way consistently for more than 60 days straight. I had never hit the limit even once. Over the past week, that has clearly changed.
In previous weeks I was running 2–3 projects at the same time with much heavier workloads, and the limit still didn’t drain this quickly. My workflow and project structure haven’t significantly changed—the amount of usage consumed by essentially the same work has.
I’m specifically talking about the last week. After raising the issue, I also noticed a noticeable increase in similar complaints on X and Reddit. Interestingly, it seems to have become much more visible after the reset-bank sale news.
Whether it’s a bug or an intentional change, I’m convinced something has changed.
Intuition is telling me that the slow shrinkflation of tokens on GPT is premonition of it winning the market from other AI developers. I’d advice every developer to concentrate on making profitable stacks, because it looks like at some point in time consistent usage of agents “just to help me code” will be ancient history because every answer will be PAID. I hope the competition continues at least for 3-5 more years so that simple users like us can take advantage of the big boys trying to pull the blanket in between themselves.
With that being said, please stop token shrinkflation!
I agree, but this is a horrible strategy for OpenAI because they haven’t yet monopolized the marketplace. Amazon is the case study for how to monopolize a market BEFORE gouging external stakeholders.
I’ve had the exact same experience and came here to see if I was not alone.
I used Codex 20x for months. It used to be so hard to hit my limits that I would turn on extra loops just for experimentation.
Now I can’t even do my regular work for a couple of days without hitting my weekly limit. Experienced it in the past couple of days for the first time.
The bright side is that this finally gave me the motivation to seriously experiment with open source alternatives.
This is only my hypothesis, and I still don’t really understand how it works. But I’ve usually noticed it happening in waves.
My hypothesis is that it has something to do with limited compute resources. Usually, shortly before new models are released, the limits seem to drop sharply, or even the quality of responses seems to suffer. Maybe at that point more resources are being allocated to training the new model, which could also explain why we sometimes feel a noticeable improvement once the new models are released.
But maybe I’m just a freak 
I’m doing the exact same thing.
Hopefully they will become aware of that via this thread, or maybe this frolics don’t affect subscription base yet, I am in the dark about it. What is clear is deterioration of the service price-to-benefit ratio with the last 2 updates.
I’m getting almost 10X cost savings running the exact same model through JetBrains AI Assistant. That’s crazy…
Really, the quota is being consumed extremely quickly. Even a very small task can use up a lot, and I am not sure if there is a new bug。
Normally, I use 5.6 Sol, and previously the consumption was very low. But now the consumption has skyrocketed. Excuse me, has the floodgate been opened?
Yes I also noticed the quota on Pro Lite is being consumed super-quickly as compared to a week ago. I am too confirming that weekly usage limits have been halved or less. This is ridiculous.
This is happening to me as well. I’ve a Pro 5x subscription and I’ve even lowered my usage. Before I was using gpt-5.6-sol max for planning and xhigh for implementation.
Now I’m using xhigh for planning and high for implementation. The usage is decreasing much faster than before.
Same project. Same tasks size.
This is either intentional (to see how much we can cope with limits) or a well visible bug.