That was also said yesterday but yet there has been nothing done about it.
2 Likes
OpenAI, we’re stuck in a limbo here. I wanted to give you guys some time to address this, but it has been days now… And still no fix? Codex is completely broken right now. Seems like you’re pushing us to Claude, is this intentional?
4 Likes
I don’t know what OpenAI did in the last 2 or 3 days, but they completely ruined Codex, the only reason I was paying for it. I hope this is a bug and not an intentional change.
2 Likes
We’re also the best users you want to train your model so shutting us out slows the entire Codex development down. @OpenAI_Support is there anything we can help you with to solve this faster, seems like the community has already diagnosed the issue AND offered multiple pathways to resolve this now, let us know if you need data, etc.
If it’s about the refunds I’m sure the users understand if they are processed separately, the goal should be to get the platform back up ASAP, fix the token issue or rollback back to the older model? You’re the top model now but if you stop progressing Grok and Elon will steamroll you soon, they’re aggressive and ruthless as @sama (is this Sam’s account? Sorry if the tag is wrong) is aware.
We’re here to help if you need it, but some transparency would go a long way and would be greatly appreciated. Having an active dev community should be your strongest asset, harness it to your advantage.
Hope you have a great day, and thanks.
2 Likes
Same here, the codex on plus plan is a joke now. Its like they slashed the limits by 90%. Codex is the only thing that kept me subscribed to plus, if this new limit is not a bug, i cannot see myself being subscribed any longer.
I wanted to share a concrete example of what I just observed Codex achieve and what it cost on the usage dashboard. To my mind, it highlights why the current usage model feels structurally unsustainable even for moderate users.
I ran one short Codex session on my project.
All it did was:
-
Open one Python file
-
Apply a small, deterministic patch ( +8 / –4 lines )
-
Commit and raise a PR
-
Pass CI hygiene (
ruff,mypy --strict, I explicitly told it not to runpytest)
Total Run Time : 1 minutes 29 seconds. Then I pushed the PR through to GitHub.
-
The GitHub connector automatically triggered the PR Review.
-
The total runtime for the PR Review was was about 2 minutes 37 seconds.
No multi-file refactor, no test generation, not test execution, no documentation synthesis, literally one surgical fix of tightening a test guard.
When I checked the dashboard afterwards:
-
5-hour usage: 99 % remaining
-
Weekly usage: 3 % remaining (from 4 % yesterday)
-
Credits: 0
In other words, that single two-minute patch shows as it consumed ~1 % of my weekly quota.
That translates to roughly 100 credits for one clean, CI-compliant change, the smallest possible “professional” workflow. If that’s 1 %, then a user doing five such loops a day (which is normal for any active developer) would exhaust their allowance inside a week.
The problem isn’t just pricing, the problem is structural:
-
Responsible engineering is credit-dense.
Every step OpenAI encourages using Codex to enhance and support professional development work - that, in practice means, running tests, linting, static analysis and CI validation and these all count as a “task”. -
Predictability is opaque.
The percentage bars don’t convey what one session costs in practical terms, making planning impossible. -
The model penalises good practice.
Cutting corners (no tests, no reviews, one-shot patches) would be cheaper than doing things properly.
Codex is genuinely extraordinary, it wrote this patch flawlessly, respected the AGENTS.md governance rules and merged green. When I ran pytest locally - no issues - perfectly clean.
But when “doing things the right way” costs 1 % of a weekly limit, it discourages precisely the kind of disciplined, reproducible engineering that makes Codex shine.
If the answer is simply “upgrade to Pro and buy credits”, that’s not a sustainable path forward.
At current rates, anyone using Codex properly, running tests, generating documentation, validating CI and shipping production-grade code, could easily end up spending around $1,000 USD per month once you combine the Pro subscription with credit top-ups.
That might be feasible for an enterprise, but it prices out the very people who can help Codex evolve, independent developers and small studios who are actually pushing its limits responsibly.
3 Likes
yeah, easily could run up 1000$/month… I managed to use $5 in credits over the course of… looks like 1 hour… This is insane.
It should not cost this much.
3 Likes
this is getting ridiculous man i just started using legit did least work went from 100% weekly and daily to 80 daily and 87 weekly how are u consuming credits just for writing few lines code and search this crazy, if its a bug fix it , border line not unusable at this point of time
2 Likes
Fix the bug please! this is insane!!
1 Like
Same here.
It’s really getting frustrated when Codex replies “I’m sorry, but I can’t help with that.”, yet charges you for the result.
Then I have to re-run the same prompt multiple times for it to actually produce the expected result.
Complete robbery.
3 Likes
-
I’ve ended my day yesterday with 70% 5h limit and 57% weekly limit.
-
Woke up today with 100% 5h limit and 33% weekly limit.
-
After trying to submit a new prompt, my 5-hour limit dropped to 99%.
-
I entered new prompt and got a 97% 5h limit.
I’m not sure this is working correctly. 
3 Likes
It absolutely isn’t working correctly. Bugs happen, I get that (this one seems like it should have been easy to pick up given how many people are affected but w/e). And fixing them may take a while. But rapid mitigation is surely possible? For now they should suspend limits or proactively and rapidly respond to requests to have them reset. @OpenAI_Support this isn’t acceptable.
2 Likes
I upgraded to Pro, which was not so straightforward since I was on a business account (a new account had to be created). For Plus/Business, Codex is no longer really feasible.
Ofc, with Pro, you can do a lot more with Codex. That said, the weekly limits are still very, very low (enough for three days of decent, but limited use). Even with Plus you could do more before the changes. My suggestion: While I understand that more income is important, @Sama, I think developer plans would make more sense for users who do not need the ChatBot, which is kind of unlimited (which is somewhat ironic since the model is pretty much the same as Codex), or Sora, which is likely expensive for OpenAI.
Food for thought.
Another thing, Codex many times simply does not want to follow the promt when it considers the tasks too large and on cloud you often see these “I’m sorry, but I can’t help with that.“ messages. This should not count towards any limit.
I will wait and see how this turn out this month.
Same here. I came to check if anyone else is having the issue and it seems it is not just me!
1 Like
Just saw that following a brief Codex outage the rate limits have been reset:
1 Like
I noticed that you reset the limits… great! However, I’ve already used 22% of my weekly budget and 74% of my 5-hour limit with just two messages in a single cloud task today, and that wasn’t even including a code review. The usage has been terrible since November. Honestly, I wouldn’t mind a daily 30-minute outage if it meant the limits would reset properly.
At this point, it’s become unusable. I’ve switched to another cloud platform until this gets sorted out. I’ll probably just use it occasionally for a code review, because Codex isn’t usable for anything else right now.
See that tiny 1-pixel blue line? Yeah, that’s my Plus usage today. Even if Pro increased it by 10× or 20×, it still wouldn’t come close to the pre-November levels of coding.
3 Likes
They managed to solve that issue quick, but we are still waiting on news for the bug with usage.
I love that everything is reset, but when my weekly usage is gone in only a few prompts it doesn’t really do much.
1 Like
Once I saw the notice on X and the credits reset, I decided to run something close to a control test to understand how my work actually affects Codex’s usage and burn rate.
Sequence
Prompt: single structured request — a standard multi-file plan → patch → PR creation (i.e. normal workflow).
Codex output:
• Two files modified (+15 / –3 lines total)
• ~10 minutes of active Codex processing time
Automation: On PR push, the GitHub → Codex connector automatically triggered a code-review job (~2 min, one shell command).
Total observed consumption
-
5-hour limit: 21 %
-
Weekly limit: 6 %
That means a single 12-minute end-to-end cycle consumed the equivalent of ~66 minutes of quota — roughly a 6× multiplier over real compute time.
| Stage | Runtime | Approx. quota used | Notes |
|---|---|---|---|
| Prompt → PR generation | 10 min | ≈ 63 min (21 %) | Multi-file patch |
| PR review (auto) | 2 min | ≈ 3 min (1 %) | Triggered by connector |
If that scaling holds, a single engineer doing 8–10 normal PRs per week would exhaust their allowance within a day.
From what I can tell, it’s not token burn, it appears to be event-weighted (or possibly both). If it is token burn, then there’s no practical way to forecast consumption, because there’s no visibility into how a given prompt translates to a token-cost equivalent.
For context: this was me working on a single-repo session, not working on multi-repos in parallel having a number of prompts processing simultaneously.
I could easily have up to 10 prompts in parallel across environments because that is the force multiplier that Codex presents - that’s the productivity opportunity it presents, which under this model would scale the burn roughly linearly across each concurrent task, easily consuming an entire weekly limit in a single session.
If the only remedy is upgrading from Plus to Pro (10× cost) and still buying credits on top, that’s an effective outlay of around $800–$1 000 AUD/month just to sustain regular development on one project alone without even considering working on multiple projects.
That’s simply not viable for solo developers or small teams.
Right now, I’ve solidly formed the view that the Codex pricing and burn-rate model are economically unsustainable for continuous development workflows.
3 Likes