Codex Rate Limits Discussion Thread

Codex
147

I’m happy to report that my hypothesis was most likely incorrect. It was probably just a coincidence, a fluke.

Because I just logged back into my account, and everything started working properly again. I asked a couple of questions, and my limit didn’t even drop from 100% to 99%. This suggests that everything is working just as it did before. And perhaps OpenAI has already fixed the bug that was there. I’m incredibly happy that I can return to my old accounts and continue working on my applications.

1 Like
148

Thanks for the heads up. I haven’t done any coding or anything today yet, but I hope I to experence the same. Will report back after lunch.

@Mat_Tr Your right, someting has changed withinn past 24 hours. I am still eating credits faster then we used to, but thats to be expected. Its now more usable, not as much as it once was but still usable nevertheless.

Update
While it is better, or seems so, still burning too quickly to use on any serious coding project.

1 Like
149

Hello,

I want to provide direct feedback based on my current experience using Codex CLI.

In its current form, this setup is not acceptable for me.

I have already spent a significant amount on API usage (in the range of several thousand dollars) and I am also subscribed to ChatGPT Plus. Based on this, I expected Codex to operate in a predictable, usage-based model tied to billing.

Instead, I am being blocked by a separate weekly Codex limit that:

  • is not clearly tied to my actual usage or cost

  • cannot be extended or managed in a transparent way

  • interrupts active development work at critical points

From a user perspective, the system is extremely difficult to understand and plan around due to the separation between:

  • ChatGPT plan limits

  • Codex-specific limits

  • API billing/credits

These systems behave independently, but this is not clearly communicated in practice. As a result, even with available credits and prior spend, I am unable to continue working.

Additionally, the recommendation to switch to smaller models is not a complete solution:

  • Smaller models can introduce inaccuracies or incomplete changes in complex tasks (e.g., refactoring)

  • This leads to additional iterations, corrections, and repeated requests

  • In practice, this can negate the expected efficiency gains and still consume limits quickly

At this level of pricing and usage, I would expect:

  • a transparent and unified usage model

  • predictable scaling based on billing or credits

  • or at least the ability to control or extend limits when needed

Right now, I have invested significantly in API usage, but cannot reliably use Codex due to these separate, restrictive limits.

This makes Codex difficult to trust for real-world development workflows.

My questions are:

  1. Is there a way to run Codex purely on a billing-based model without hard weekly limits?

  2. Can these limits be adjusted or increased for accounts with significant API usage?

  3. What is the recommended configuration for uninterrupted development workflows?

If there is no clear solution to these issues, I cannot continue using Codex in its current form.

I would appreciate a clear and transparent answer.

150

I apologize. I started celebrating too early, which may have created false hopes. After switching back to my previous business accounts, the initial requests didn’t lift my limit, but it eventually happened with a delay—and still in an unreasonable volume. I believe this delay in the limit update is related to the new token-based request pricing system that went into effect for business accounts on April 2nd. For example, on a Plus subscription account, these changes haven’t been implemented yet, and you can see limit updates immediately, without any delay.

I am truly leaning toward what you mentioned earlier—that such rapid token consumption is primarily due to the new token-based pricing system that took effect on April 2nd for Business subscription holders. This is exactly why everything still works as before on the Plus account. However, the thought that these changes might soon affect other plans as well frightens me, as it would make Codex completely unusable.

I am at a complete loss. I have no more hypotheses left. My only hope is that OpenAI will recognize that the request pricing they applied to Business holders is unjustified and unsuitable for using the tool, and that they will make significant adjustments to its operation, making the interaction process fairer and more practical for work. At this point, I am suspending my activity in this thread until there is a clarification or resolution of the situation from OpenAI. Have a great day, everyone.

1 Like
151

@Mat_Tr Likewise Mat, I think we have clearly made case and point.

1 Like
152

To be honest, I think providing a simulation of peak consumption under the new pricing model before the change takes effect would be very helpful for customers. I recognize that the previous business rate was quite generous, and I’m personally fine with an additional €80 (this is what I pay for claude) on my monthly subscription to maintain the same capacity. However, I currently lack the data to determine if that will be sufficient compared to my previous rate.

153

I think we should all just wait till OpenAI clarifies one simple thing - is it a BUG or NEW POLICY?

2 Likes
154

I’m honestly not sure why people here are trying to hunt for a “bug” - there is no bug. Nothing is broken. This is exactly how it was designed.

OpenAI cut Codex rate limits for Business accounts by around 60% compared to Plus. That’s not a theory, it’s right there in their pricing. Pretty hard to miss.

And no, this isn’t some mystery or oversight. It’s very obviously intentional. First, limits get slashed. Then suddenly and how convenient no-monthly-fee Codex seats show up. Not as some generous perk, but as a way to funnel usage into paid API credits instead. You want real usage? Great, go buy it.

This isn’t new behavior either. Anthropic did the exact same thing with OpenClaw, killed native support and now, surprise, you need to use their API for it to work. Funny how that keeps happening.

Let’s not pretend this is anything other than a monetization shift. Every AI company needs revenue (and especially wants to show it before IPO conversations start heating up), and who better to extract it from than businesses?

And sure, big established companies probably won’t care. They’ll just pay and move on. But for smaller teams or companies still scaling, this absolutely matters especially if you’re already locked into annual plans.

Because yeah, it does feel ridiculous, you pay more for a Business plan and get significantly less included usage than a cheaper Plus plan.

AI is clearly heading toward commodity status, so we can probably expect more of these “surprise optimizations” down the line. But this one? Flawless execution: zero warning, zero transition, just a blunt “don’t like it, then leave or start paying. Your choice.”

2 Likes
155

Wait, a HAR file from codex? How is that possible, who runs codex in a browser? Am I missing something, how does one do that from Codex CLI or the Codex App?

1 Like
156

just wanted to say thanks for the reset, Tibo

1 Like
157

Yeah, it looks like it happened a few hours ago, but I didn’t want to mention it in case it hadn’t rolled out for everyone.

:folded_hands:

158

@d_ul Good observation on the differences between Plus and Business. I would just add:

This isn’t quite the same thing? OpenClaw was a third party tool (to Anthropic) over which Anthropic had no direct control and the usage of the API with bundled Auth credits for a different set of services feels a little on the exploitation side to me.

In this case, of course, OpenAI is dealing legitimately with its own product and product limits.

Here in the UK Business plan is slightly less per seat than Plus and of course has several additional elements to the service, which are worth bearing in mind. But the differences in bundled credit is interesting.

1 Like
159

The issue still persist, on business accounts. @OpenAI_Support any updates whether it’s BUG or new POLICY?

1 Like
160

personally i’m not upset cuz i have my own local LLM but just for example sake i have burned through 30% of my weekly usage in less than 5 hours limit lol and also the task i set to do hasn’t been solved, this is rather pathetic on codex side

161

hmm … it’s just that it’s not a free lunch anymore and they need to pay for the processing of tokens you are using, pay back creditors and take a reasonable profit.

I suspect this won’t be the last price/service level change.

162

i mean the fact that the task was not solved, i also got a ‘model has reached capacity limit’ on 5.4 and switched to 5.4-mini which had to finish the logical work, overall i’m moving away from gpt but i do want to keep using codex cli

1 Like
163

well that’s a different Topic … personally I find Codex very capable and its night and day from 1 year ago …

164

This topic is about another thing, not about defending/attacking codex capabilities. It’s about 5hr LIMITS that have been dropped significantly (tenfold). I’m happy with week limits on business, but with 5hr limits I can’t use noew even HALF of week limits. Business works 8 hrs per day, not 24 hrs. Think about it @OpenAI_Support please!

1 Like
165

Small update: for what it’s worth, by asking Codex to abide by the AGENTS.md file where I clarify that it has to be extremely strict with tokens and communicate as little as possible, it’s significantly reduced my token use. I feel like 5.4 is just way too chatty in IntelliJ Idea by default. A great means of “giving people more usage” is to just give us the option to decide how communicative we want it to be. I don’t need a paragraph for every single change it makes, or an explanation for everything it’s doing along the way. It’s a win/win for users and OpenAI.

166

For context, I have been using roughly 200 lines of rules in AGENTS.md files for a while now, but the limit is still being blown out of portion on business accounts. Pritty sure tho without them it be far worse then it is.