Codex Rate Limits Discussion Thread

Codex
212

I’m not complaining about them cutting the weekly quota in half. The issue is the 5h limit. It breaks continuity. That is a huge problem for those of us with pipelines that have used codex for only one reason: because it can maintain continuity for long conversations.

The irony is post June I’ll be scaling and will be using usage based anyway. I shouldn’t care, right? Well, April is very bad timing for them because I am not ready for that during this incubation period. They are forcing me to replace Codex. If they delayed this for 3 months, they’d be enjoying a lot of revenue from its place in my pipeline as it scales. But, because they did it now, they are forcing me to optimize today. That’s likely not reversible. I won’t undo the optimization in June.

So, I’m not here to save myself. I’m here to try to give OpenAI a warning that they are causing too much friction way too soon in the hopes of increasing usage based and it will likely hurt them in the long-run. Once I replace Codex, I’ll never go back. I’ll already have a pipeline that no longer needs it and be optimized for lower cost. Today, for the first time, I already replaced Codex for most reviews. I need continuity during execution of an implementation plan above all, and today, I no longer have it from Codex.

On top of that, I have to make a decision on the annual renewal of the business count by next month.

213

Again, I was aiming my comment about the many, very slanted comments I sorta wish I hadn’t read.

I’m talking to you because you seem pretty reasonable and pragmatic, which are qualities I respect.

Generally, though, what are any of us to do? This isn’t like Burger King or the local cafe where we can leverage our business with any real meaningful impact….

I might not have the full picture of what’s going on, and I can only extrapolate from what reaches the public… but it seems like OpenAI is focusing on coding as its baseline, as being an open-armed chatbot proved to be very dangerous for litigation… (though little fault of their own in my opinion)

I understand the costs are gated differently. I can see that the codex models are very proficient at what they do compared to the general purpose that is public-facing…

But, I don’t have enough data points to make a certain, clear opinion on the pricing tiers for Codex…

I’d like us all to be successful… we live in an age where none of this was considered possible even a handful of years ago.

Maybe I’m too humble about it, maybe I’m a sucker… maybe I’m a fool…

I’m grateful for what this company has put into the world, and I am sorry about your workflow… I really do wish you the best, and generally things work out…

You seem like you’ve got that ‘get it done’ attitude that you need to survive the turbulence…

I just hope that things work out for us all.

:clinking_beer_mugs:

214

I think I made my decision regarding annual renewal - NEVER AGAIN 100%.
You can’t have your business dependable in any way on this company. Period.

Occasional usage with occasional monthly renewals - very likely yes, it’s a great product for sure, nobody has any doubts I think.

A great example where delusional management destroys great product.

215

I’d rather not dive into the difference between data protection difference between a user and a business, but what I said is still there:

You can at anytime use a plus account to run Codex. Obviously if you feel comfortable running the data through a personal account, then it’s a great choice.

Probably for the best. These companies have changed direction and policies so much that any long-term commitment is probably going to be a bad time :rofl:

216

I’m on the Plus plan. My Codex usage is about 60 prompts per day and 6m tokens a day. I use GPT-5.4/medium most of the time. I’ve hit a weekly limit once and a 5h limit once, I get pretty close to the weekly limit most weeks.

But I’m aware enough of the limits that I try not to have non-codebase-related conversations with Codex. If I want to dive deeper into why something works a certain way, I use ChatGPT instead, which is friction.

Something I’d like is for the “Usage” readout to be time relative. Seeing “48%. Resets on blah” is somewhat useful if I get my calculator out, but I really want to know if I’m on track to go over or under. I think something like “5% over budget” would be the sort of stat that’s confusing at first but after many peeks it would become just what you want to know.

217

I don’t know… Codex seems pretty chill regarding rate limits via web-app…

image
image1179×2556 215 KB

218

I’m genuinely surprised by the people commenting in this thread without having a Business account and without properly understanding the issue, yet claiming that the problems reported by Business users are “normal.”

The fact is, they not only don’t use a Business account, but also have only a superficial understanding of what’s actually being discussed here.

Everyone knows — including Business users themselves — that since April 2, the Business plan has become roughly 60% worse than Plus. It’s also common knowledge that the “double limits” promotion has ended.

But that’s not what people are complaining about.

If the Business plan had simply become 60% worse, as stated, that would be unpleasant but acceptable. Even if it were 100% worse, I could live with that.

That’s not the issue.

Users are reporting that limits are being consumed abnormally fast. We’re not talking about a 60% difference — we’re talking 500% or more. At that point, the tool becomes unusable.

Meanwhile, outside observers seem to assume they know better, concluding that users simply don’t understand how to manage their limits.

A simple example: I mentioned that a request containing just the word “hello” could consume up to 5% of a 5-hour limit. The response I got was that I must be using the system incorrectly — instead of considering the obvious possibility that something is wrong.

For clarity: this request was made in an empty project, in a new thread, with no prior context and no agent instructions. Clearly, it should not consume that much.

Because of this issue, I switched back to my Plus account. Yes, limits there are also being consumed faster than during the promotion, but still within reasonable bounds: over 5 hours of active work, I use about 50%.

On the Business account, however, the same workload drains the entire limit in about 10 minutes. That’s not a 60% difference — that’s a critical degradation.

As for the cause, it likely relates to the new pricing system introduced on April 2. Usage is now calculated based on tokens rather than message count. For now, this applies only to Business accounts, but it’s expected to roll out to Plus, Pro, and Enterprise plans in the coming weeks.

If this system isn’t refined, these same issues will affect everyone.

Bugs in newly released systems are normal. That’s exactly why we’re raising this issue — so developers are aware and can address it as quickly as possible, or at least clarify whether this is now the intended pricing model.

So I ask those who haven’t encountered this issue not to argue or derail the discussion. It only makes it harder for the team — who have already received these reports via the moderator — to properly review and respond to user feedback.

219

Em dashes are rough, eh? Just saying…

… mat linking mat?… just saying…

221

After today’s Codex Desktop update on macOS, it feels like token usage on Business accounts has become somewhat more reasonable compared to before. It now seems closer to roughly a 60% difference between Plus and Business.

That said, I’m not entirely confident in this conclusion yet, since I’ve only tested it with a few prompts. Still, the fact that several requests together only used about 5% — whereas previously a single request could cost that much — does seem to indicate an improvement.

I’ll share more detailed observations later today after a more active work session.

222

Just updated Codex win app, and it seems/feels like they probably solved the issue, 5hr limit feels better and usable. Here’s changelog. Oh yes, it says nothing about update…:melting_face:
@OpenAI_Support Why ain’t you guys using Codex to push changelog website update, it’s so easy, I can share Skill if you need!

223

I can confirm my earlier statement regarding Business accounts. The volume of requests that yesterday would hit the five-hour limit within 10–15 minutes can now be handled for about 1.5–2 hours at the same pace.

If yesterday the Business plan was practically unusable, today it’s actually workable — not quite like before, but still noticeably more fair toward Codex users.

Those who said this was how it was supposed to be were wrong — even though they were acting like know-it-alls :slightly_smiling_face:

In any case, it looks like we got the result we wanted — even if the fix rolled out without any official announcement or response from support in this thread. Wishing everyone the best and success with your projects!

224

looks like rates were reset too…

I appreciate it, thanks team!

225

The Pro X5 subscription has just been released, offering 5× the Codex usage compared to Plus. But that’s not all — there’s a promotion running until May 31 that boosts it to 10× compared to Plus!

That’s a seriously big update. For $100, getting the ability to practically forget about limits for two months while using Fast mode sounds very compelling.

Looks like I might have to put my investments in Business accounts on hold for now and invest again — aiming to get the most value possible out of these two months.

226

A strong suggestion with the new codex model update that dropped today ~

It’s more than a rate limit reset, they upgraded the version…

If you have projects you’re working on, spend the time to have your codex examine the project architecture first, let it explore its working environment…

Those two sorts of passes will make the transition easier, as far as I understand things.

227

For those that weren’t clear, please don’t miss the useful explanation of the Business plan here:

Key point being adding seats does not add codex credits, you just get new access to the central workspace pool which you need to maintain with additional purchased credits as required.

228

Do you have Business account? This is misleading. Adding new seat in business gives you credits (limits, not really credits) for both chatgpt and codex, independent of credit pool. You don’t need credits if limits are enough for your seat. The subject title of this topic says “…Limits…“

229

Are you saying the OpenAI staffer is being misleading? I would reply to them directly if so.

230

He doesn’t lead to misleading. He speaks not on the topic that is discussed in this topic.

231

I have no idea, I’m just saying what I have in business acc. You buy additional “normal” seat and you get chatgpt+codex. That’s what I did 6 times. That’s what they also promised when I bought it. See, you get [NOT]“Everything in Plus“ and “standard or usage-based Codex seat“ which many believe here is “subsidized“ [by Soros I guess :rofl: ].

image
image628×621 37.9 KB

232

Just linking and paraphrasing, feel free to take issue with the author.