Codex (gpt-5.5, Plus plan) — rate-limit cost per token jumped ~10-20x since June 16, draining the 5h budget in 2-3 prompts

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #28823
• #28498
• #28727
• #28065

Powered by Codex Action

Same here. I'm on the Pro plan and my quota suddenly vanishes in less than 5 hours now. Even with top-ups, it doesn't survive a single intense work session. This rate-limit spike completely breaks the entire workflow. Attaching my restriction message as well (referencing image_f0485a.jpg / my previous logs). Hopefully, this gets fixed ASAP.

+1 here,

seeing a sudden and unusually large drain in my Codex 5h primary limit.

After a single simple conversational interaction, my 5h limit dropped from approximately 99% remaining to '67%'.. =(

This seems materially higher than expected for the type of interaction performed.

(Environment)
Plan: Plus
Surface: Codex Cli
OS: Linux
Model: gpt-5.5 high
Approximate time: 2026-06-18 22:20~30 KST

Same here. Even more than that - after a few or a dozen prompts throughout the day, without any large documents (at most a dozen or so kilobytes) and without using MCP, my profile shows, for example, 22-26 MILLION (!) tokens used. And I have reasoning set to Low or, at most, Medium. This is crazy! At these prices, if I had to pay for tokens, I’d be paying for the 5.4 model (since I don’t even use 5.5) at $15 per million - that’s over $300 a day! I think they’re being calculated incorrectly. This can’t be accurate information about token usage.

I had to switch to Claude; now Codex seems like a more aggressive Claude 🤣. Now I feel like Claude doesn't use any resources. Codex is unusable today.

I'm experiencing the same behavior on a Pro 20x account.

Since approximately June 18, my 5-hour limit has started draining at a rate that is completely inconsistent with historical usage. The weekly quota is decreasing proportionally as well.

Before this change, I was never able to exhaust my 5-hour allowance, even during heavy Codex usage sessions involving large codebases and long-running tasks. Since yesterday, the same workflows are consuming the entire 5-hour budget in roughly one hour, which is much closer to the behavior I previously observed on a Plus plan.

Nothing has changed on my side:

• Same account and subscription tier (Pro 20x).
• Same workflows and usage patterns.
• No significant increase in prompt size or task complexity.
• No changes to models or configuration.

The timing closely matches the reports in this issue and other recent quota-related reports. From the user perspective, it appears either:

1. The effective weighting of usage against the 5-hour budget changed around June 16–18, or
2. The quota accounting system is currently overestimating usage.

Given that multiple Plus and Pro users are reporting the same sudden regression over the same time window, this looks more like a platform-wide accounting or rate-limit issue than a change in individual usage patterns.

Would be helpful to know whether any quota calculation changes were deployed around June 16–18 and whether the current behavior is expected.

Same pattern here, from a paid Pro/Codex/Hermes setup. I posted the detailed trace in #28823, but the short version is:

2026-06-19 19:06 AEST backend endpoint:
primary 5-hour used: 52%
secondary weekly used: 59%
weekly reset: 2026-06-25 08:46 AEST
plan_type: prolite
Spark separate bucket: 16% weekly used / 0% 5-hour used

This is less than one day into the weekly window and after only small recent interaction counts. OpenAI’s public Codex pricing says Pro is “5x or 20x higher rate limits than Plus,” while Plus GPT-5.5 local messages are published as 15–80 / 5h. The practical behavior now feels materially inconsistent with that representation unless there is an undisclosed reweighting/multiplier or mis-accounting bug.

The product also pushed an “upgrade because you are nearing limits” style message, which is not an adequate answer for an already-paid plan when the issue is unexplained quota drain.

What users need is a per-turn/per-feature ledger: model, client/surface, bucket charged, input/cache/output/reasoning tokens, tool/MCP/image/retry/startup/compaction/background charges, and any multipliers.

Detailed trace: https://github.com/openai/codex/issues/28823#issuecomment-4750189876

+1

Same here on a ChatGPT Pro / Pro 5x account.

I primarily use Codex through the CLI. Over the last ~2 days, the 5-hour usage window has started draining much faster than normal, despite my workflow being basically the same as before.

Same issue here. Saw this start to happen when the 2x offer ended, and the remaining was not half.

This happened to me today as well. Curiously, right after I started working, I noticed my tokens were being used up very quickly relative to the five-hour usage window. I'm currently having the Pro plan.

Does anyone know what might be going on?

+1 to this, I'm a power user in Mac. I'm seeing this since June 1 drastically increase in rate limitsvl consumption. I'm on 100 dollar plan and I see for 5 hour session I'm usually finished the limit in 1.5 hour majority of the times

Does anyone know what might be going on?

I've got a feeling this might have something to do with Codex memories - I speculate that they might have upped the context being fed with each request or something related. Codex (and ChatGPT) seems to be overly reliant on memories the last couple of weeks.

Last night, I tried attaching the following prompt snippet to the end of my prompt to solve a really complex task which I had been struggling with for days:

"Don't rely too much on using memories and past implementation notes. Use your own solid judgement and understanding of what the best approach would be to achieving the desired outcome"

I was using gpt-5.5 Extra High reasoning. I didn't check if it used less tokens, but it definitely did a brilliant job of accomplishing the task - with no bugs left in its traces. Seems to have as much of a model performance improvement as what a fresh context conversation gives you.

However, there are definitely still cases where you will still want to be utilizing memories for what you are doing though. An enable/disable Codex memories toggle would be really nice to have at the top right corner of Codex Desktop, since the user currently has to go deep into the settings dialog to toggle it on/off.

I thought I was alone.

I'm on the go plan and was keeping codex quota saved for a heavy lifting session.
One read file tool call and a response is all it took for my entire monthly quota to vanish!
Yes the entire month, not weekly nor a 5 hour limit, it refreshes next month. 💀

+1 here

I have the same issue.

Same here, running a goal for 15minutes drain all 5 hour limit. Im using Plus account

Its not just gpt5.5 , i had tested kimi2.7 and token usage went through the roof, not getting even 1 hour of work before 5hrs limit kicsk in, on a bigger codebase, and glm5.2 connected to cursor, gives a little more usage then kimi an its cli, but not by much, but cumoared to this two, gpt5.5 still gives more usage

Same issue i am facing with my plus account it barely survives 7-8 lomg working prompts, for just creating a dockerfile and running it, it consumed around 40-50% of usage limits on my plus plan with gpt 5.5 medium reasoning and the prompts were detailed

I thought I was alone. I'm on the go plan and was keeping codex quota saved for a heavy lifting session. One read file tool call and a response is all it took for my entire monthly quota to vanish! Yes the entire month, not weekly nor a 5 hour limit, it refreshes next month. 💀

Same here

Same here. Where normally I would work 3-5 hours a day, now in just 1 hour I'm done. Also in roughly 2-3 days I've went through 85% of my weekly quota. I will definitely give up on Codex. This is a scam. Imagine you pay a monthly subscription only to have the rules changed midway, no notice, nothing. I will go back old school where I write my own code, with no security issues.

Same here

I think this is an absolute disgrace for OpenAI

Yep, same here. Thought it was something I did when a single prompt yesterday took like ~70% of my 5h limit (Plus plan). Glad to see I'm not alone. Do I smell a new reset incoming?

Same here.

+2 same here, from 2 pro 5X accounts

same here

Has anyone heard whether this is a temporary issue or a permanent change? I'm experiencing the same behavior since around June 18. Previously, I could get 15–20 prompts every 5 hours on gpt-5.5, but now the quota gets exhausted after just 2–3 prompts. I'm wondering if OpenAI is aware of this or if anyone has received clarification.

same here

Finally! Some real stats! And yes i feel this too and it’s getting hard to work with codex lately and making me move. Hopefully this gets fixed

Codex 5.5 is not usable at at all right now. Facing the same issue. Is this a bug or intentional? We at least deserve some explanations.

it's not even usable anymore

+1 Here.

It's not usable for any task now.

Also it's so dumb

same here! canceling my subscriptions!

I got same issue.
I think in a few days It could actually one-shot the limits just like claude

Same, I run two paid pro accounts, and I've tried to streamline agents.md -> doc review, sectioning prompts into smaller amounts, limiting the length of chats before moving to a new chat etc. I am using codex gpt 5.5 high.

I'm hitting limits left right and centre that I'd never hit before using way less.

They 100% changed something and didn't tell anyone. I cancelled my plan until this is resolved. I'm not going to give money to a company that will just change something behind closed doors and not tell anyone using the service.

+1

+1 I am on $200 plan and weekly limits got consumed 30% in just one day. This is ridiculous.

Same here on Pro plan.

+1

+1 same for me with 200usd pro plan, gpt subscription makes no sense, claude calls me deeply...
gpt 5.5 got dummer model ever for last 2 weeks and it's not joke...

Same here

<img width="1030" height="822" alt="Image" src="https://github.com/user-attachments/assets/f605907e-dc2d-4b90-b337-a3dbbe7ae64f" />

I thought this was some conspiracy or collective delusion, but the drop I’m experiencing is brutal. I used to be able to use it for at least 2 hours straight. Now, in the last 24 hours, I’m hitting 10% after just 30–40 minutes.

+1

Same here, im planning to move to Cursor, Antigravity or Jules,

Same Issue, I used to have much more done but now one prompt worked for about 15 minutes and consumed 20% of the weekly quota, and the entire 5h quota.
like this even if 5.6 if efficient we won't be able to use it with these limits.

Yup, my weekly quota is draining 2x(if not 3x) faster then it did 2 weeks ago.
P.S. I’ve also noticed that the speed has noticeably regressed.

P.P.S. The 5-hour quota is basically gone after just a few prompts. I used 4 prompts with GPT-5.5 and 2 with GPT-5.4, both on Medium thinking, and I’m already down to 29% remaining.

Good to know I’m not imagining it. +1

lot of good it does us to give people free resets and pump token prices 10x

Can also confirm this from my side. I was usually able to do everything I needed within the 5-hour limit, and it felt just right since I would typically have enough usage left until the next reset. Now, however, I can burn through over 90% of that same 5-hour limit with just one or two prompts.

Couldn't agree more , I do feel the same .there is something that is not transparent yet

Same here. I started using Codex a lot, but now I'm back with Claude code

The strongest next diagnostic would be an explicit "limit units charged" field alongside token counts.

Right now users can infer the regression from token_count and rate_limits, but the missing contract is the conversion step:

• input/output/reasoning/cached tokens
• model and service tier
• plan/window id
• primary vs secondary window
• limit units charged for that event
• pricing/quota conversion version, if that exists internally

That would let a bug report distinguish three cases: token accounting changed, quota conversion changed, or the UI is reading the wrong window. Without that field, every report has to reverse-engineer the same percentage deltas from logs.

---

_Generated with ax._

The same for me, I asked codex to start the server on localhost:3000 to test, I do this with codex because usually the consumption is very low (<1%), but yesterday it took me 50% of my 5 hours limit 🥲

same here, i'm on pro $200 plan

The usage in the last few days has been crazy, I'm burning through usage without any increase in output/usage. What is going on?

+1 happening to me as well (pro 20x)

Happy to find this post, I thought I was doing something wrong by accident on my end.
Around Easter OpenAI announced a pricing update for the VS Code plugin of Codex which I intensively use on a daily basis.
As a token (pun intended) of appreciation for being a heavy user, they gifted me around 10k or 11k credits.
At this time I upgraded from Plus to Pro lite which seemed to work brilliantly for a while, I rarely ended up consuming the small buffer of 10k+ credits.

Somewhat over a week ago:
ALL 10k+ tokens consumed in just 3-4 days.
I was flabbergasted and now I see that I'm not the only one apparently.

Will reach out to support for a clarification.

Update:
I requested a human review, I suggest others to do the same.

<img width="1080" height="2400" alt="Image" src="https://github.com/user-attachments/assets/9ce8f647-28d9-41a5-8aa2-cd799aebc293" />

<img width="1080" height="2400" alt="Image" src="https://github.com/user-attachments/assets/407cc102-1810-43f0-9d1f-f2f58be63178" />

<img width="1080" height="2400" alt="Image" src="https://github.com/user-attachments/assets/c9b2927f-4ef4-48e6-9027-9ef6ef3d2fe1" />

+1

just 5 prompts on Plus account then finish 5hours limt 😂

Same here

I'm experiencing the same problem.

I've noticed this problem has been around since the beginning of this week.

That's right

<img width="787" height="202" alt="Image" src="https://github.com/user-attachments/assets/ef837bba-8f5f-43d2-8408-ed9d2eb6b836" />
1 prompt gpt-5.5 and this is crazy lol

I have been having the same issue along with really poor, degraded responses (pro-plan 5x usage). My issues also started around June 16th. I had escalated to their support team, however, I received a generic response which didn't really acknowledge the fact that this problem is effecting so many people. I switched off the memory feature and that seemed to help partially, however, the usage consumption is still drastically higher than a couple of weeks ago.

I have been doing some tasks today, and the consumption seems to have shifted slightly towards normal, but let's see how long that lasts :(

Same pattern here, from a paid Pro account using Codex / openai-codex through Hermes.

The issue started for me around the same June 16–18 window. My main observation is not simply that Hermes/Codex used tokens. In one detailed audit, there was real local activity. The problem is the new ratio between local token usage and the Codex 5-hour meter.

One measured window:

Date: 2026-06-18
Window: 10:10–11:59 CEST
Observed Codex 5h meter: 100% -> 45%
Observed weekly meter: 97% -> 88%

Local Hermes/Codex telemetry:

• Sessions: 12
• API calls: 90
• Tool calls: 113
• Input tokens: 7,606,798
• Cached input tokens: 2,487,808
• Output tokens: 45,488
• Reasoning tokens: 15,718
• Total token categories including cached: 10,155,812
• Reported total excluding cached as separate category: 7,668,004

So roughly 10.15M local token categories consumed about 55 percentage points of the 5-hour Codex meter.

That implies an effective full 5-hour window of only around 18.5M token categories, or around 14M if looking at the non-cached reported total. This is materially different from my historical usage pattern.

My profile activity shows prior days above 200M tokens, including around 217M on June 12 and a peak around 240M, without the same practical exhaustion behavior. I understand profile token activity and the Codex 5-hour subscription meter may not be the exact same accounting system, but the practical regression is very clear: comparable or smaller current workflows now consume far more of the 5-hour meter than they did before.

I also found local inefficiency on my side: Hermes was carrying too much retained context and over-expanding some tasks. I am optimizing that separately. However, that does not fully explain the sudden change in effective Codex allowance behavior. The same type of work was previously possible without exhausting the 5-hour meter like this.

Current status: things seem to have partially stabilized now. It is much more usable than it was during the worst period, but my impression is that it still does not feel like the previous Pro/Codex behavior. At least now it is possible to work again, but the June 16–18 regression was severe enough to block active project work.

This looks consistent with a rate-limit/accounting regression, changed weighting, cached/context accounting change, or effective allowance change around the same period reported here. What would help most is a clear per-turn/per-feature ledger: model, surface/client, bucket charged, input/cache/output/reasoning tokens, tool/context/schema/compaction/retry charges, and any multipliers applied to the 5-hour and weekly meters.

Same issue here on ChatGPT Plus / Codex desktop app on Windows.

This is not only happening with gpt-5.5 for me. I stopped using 5.5 because the drain became unusable, but I am now seeing similar behavior on gpt-5.4 high as well.

Recent examples:

• gpt-5.5 medium: controlled simple task, about 21 seconds visible runtime, one markdown file changed, no tests, consumed around 2% of the 5h quota.
• gpt-5.5 medium: less than 3 minutes visible runtime, consumed around 10% of the 5h quota.
• gpt-5.4 high: around 4 minutes visible runtime, consumed around 20% of the 5h quota.
• gpt-5.4 high: around 1.5 minutes visible runtime, consumed around 11% of the 5h quota.
• Today: about one hour of normal Codex work exhausted the full 5h window, while not using 5.5.

This is a major regression compared to the same workflow/setup before May/June. Codex used to be very usable for me on Plus. Now normal work drains the 5h window so quickly that the product is effectively unusable.

I also contacted OpenAI Support and provided screenshots, approximate UTC timestamps, feedback/session ID, and local Codex session logs. Support acknowledged they are reviewing it as a possible Codex metering/accounting issue, but I have not received a clear answer yet on whether this is expected behavior, a rate-limit policy change, or a bug.

The important point is that this does not feel like simply “5.5 is expensive.” In my case, even gpt-5.4 high is now draining abnormally fast.

Same issue here.

There is a very big issue with this usage.
I used the Token Watch extension and checked it, and I found the following issue:
When using GPT-5.5 Medium, I can use around $13 before reaching 16% of the weekly limit and 100% of the 5-hour limit.

However, when using GPT-5.4-mini Medium, I can only use around $9.50 within the 5-hour limit, while it still consumes 16% of the weekly limit and 100% of the 5-hour limit.

So, does this mean that even if you use a lower-tier model, your available usage is still lower?

I am currently using the ChatGPT Business plan at $20/month.
---
Gpt-5.4-mini

<img width="430" height="295" alt="Image" src="https://github.com/user-attachments/assets/2010c35d-7e0d-4e15-8778-a90b32c5976c" />

<img width="419" height="71" alt="Image" src="https://github.com/user-attachments/assets/164a892b-0bc6-49a0-978e-4e419bf29b0f" />

---
gpt-5.5

<img width="428" height="225" alt="Image" src="https://github.com/user-attachments/assets/e518e393-a3fd-45bf-b6dc-1cb9cc2ad502" />

<img width="423" height="69" alt="Image" src="https://github.com/user-attachments/assets/658e6739-dab7-4f30-9766-ff086d2a3145" />

Same here. I have used my weekly usage in 2 days.

Same here, week vanished in one day with a pro account

+1. I've been doing barely any work yet hitting tons of rate limits.

Same here, I've been getting a degraded experience these last week

same here, i'm on pro $100 plan

## Summary Starting ~June 16, my ChatGPT Plus Codex budget drains in 2–3 prompts on gpt-5.5, where the same model/plan/app previously gave 20+ prompts. My Codex session logs (token_count / rate_limits events) show the limit-% consumed per token has increased roughly 10–20×, with no change on my side — prompts are actually smaller now and reasoning tokens are near zero. ## Environment - Codex desktop app, Windows 11 - Model: gpt-5.5, service_tier = "default" - Plan: plus (confirmed in rate_limits.plan_type) - Config unchanged across the regression window ## Evidence — same plan, same model, same app June 12 (working — 20+ prompts/day): | Prompt | Input tokens | Reasoning tokens | Primary limit % | |--------|-------------|------------------|-----------------| | 06:08 | 22,334 | 488 | 2% | | 06:09 | 52,161 | 1,130 | 3% | | 06:51 | 57,112 | 3,645 | 9% → 11% | | 06:54 | 79,961 | 5,125 | 15% → 16% | → ~57K-token prompt with 3.6K reasoning ≈ 1% of the 5h budget June 18 (broken — 2–3 prompts): | Prompt | Input tokens | Reasoning tokens | Primary limit % | |--------|-------------|------------------|-----------------| | 07:22:27 | 18,910 | 152 | 18% | | 07:22:30 | 20,480 | 0 | 45% | | 07:22:34 | 23,469 | 48 | 57% | | 07:22:43 | 23,631 | 0 | 68% | | 07:22:50 | 23,839 | 0 | 77% | → ~20K-token prompt with zero reasoning ≈ 10–27% of the 5h budget Per-token limit cost changed ~10–20×. Smaller prompts, near-zero reasoning, yet each consumes 10–20× more of the 5-hour budget than a 57K-token xhigh prompt did on June 12. ## What I ruled out (verified from logs) - Not reasoning effort: reasoning tokens are 0–152 today vs 3,645–5,125 on the good days. - Not prompt/context bloat: input tokens dropped from 57K–80K to 18K–23K. - Not my config: model_reasoning_effort was already high/xhigh on the working days (June 12–15). xhigh gave 30–40+ prompts on June 3, 12, 13, 15. - Not the weekly cap: secondary window is only at 12%. The 5-hour window is the one draining. ## Corroborating timing OpenAI status page shows an active incident ongoing ~2 days (since ~June 16), which matches exactly when my usable prompt count dropped from ~9 sessions/day to ~4. ## Ask Was the Plus-tier gpt-5.5 rate-limit budget reduced or the per-token weighting changed around June 16? If intentional, what's the new budget? If not, can it be investigated? ## Session log references - Good day: rollout-2026-06-12T09-07-36-019eba71-5c57-70d0-a8fa-aa999731e9ff.jsonl - Bad day: rollout-2026-06-18T10-22-14-019ed99b-ea9d-7fa1-a86b-dca4e55a8e3e.jsonl - Rate-limit data is in token_count events with rate_limits payload (fields: primary.used_percent, secondary.used_percent, plan_type, resets_at).

Same here

I am experiencing the same issue on my Codex Plus plan. It is reaching the limit so quickly. I don't know whether they have reduced the limit . But probably it is costing much. Need to think of some alternative

Yes same issue

Same here. High usage started at Friday. One task - full 5h window on plus and around 30% weekly... Where day ago I could finish 2 tasks with very high trouble shooting and quality. Something changed on their end at Friday morning (7-8AM CEST).

+1 happening to me as well (pro 20x)

+1 I am finding the exact same. (Pro plan)

+1 I am finding the exact same. (Pro plan)

+1 Experienced the same issue, I have coded with much more quota last weeks

Same on a plus plan

+1 - single prompt rinsed my entire 5 hour limit, would have gotten 4-5 out of it before. Just going to make my own agent with deepseek on a vps at this point.

I thought I was going crazy but I'm having the exact same issues. This week is the first time since using GPT 5.5 that I've had 3 or more usage limit warnings over a period of 4-5 days. Currently on a plus plan

Seems like instead of utilising reasoning during the request, it's ending early and starting a new request, which is burning tokens at a 10x rate.

Before:
Im doing this now
thinking ~10 minutes
edited files

After:
Im doing this now
edited files
Im doing this now
edited files
x10+ more responses

over and over, and over and over, it's not utilising reasoning enough or at all during the responses.

+1 same bug in plus plan

Cannot believe this, its too frustrating.

Adding my voice here as well—this is highly frustrating. Like others are reporting, a single prompt is now burning through my entire 5-hour limit. How so? Hoping for a quick resolution to this.

+1

+1
kindly fix asap, we have to work hardcore

+1

Same here, I used up 4% - 6% of my 5-hour usage limit with just one instance of context compression!

+1

+1 I thought it was an isolated issue. But it seems broader than that. I'm on Pro 5x plan 5h hours pass so quick. I burned the weekly limit in less than 2 days. That's not sustainable to wait 5 days in order for OpenAI to fix its issue.

+1

+2

+1,There is the same problem

+1, Now even the 5-hour quota of the pro 20x plan cannot support continuous programming by agents (around 3 sub-agents, nonfast).

+1, I just paid for plus and it is substantially worse. The limits feel worse than the free plan a few months ago.

I've got a feeling this might have something to do with Codex memories - I speculate that they might have upped the context being fed with each request or something related. Codex (and ChatGPT) seems to be overly reliant on memories the last couple of weeks.

This is what I thought as well. I disabled memories and the usage did seem less radical than before, however, it is still noticeably higher than the normal usage consumption I had a couple of weeks ago.

same，and i had one more problem:I didn’t do anything, usage went down by itself.

+1, I just paid for plus and it is substantially worse. The limits feel worse than the free plan a few months ago.

Same here.. I regret that I paid. Is there any way to cancel the subscription and get money back?

Same. Is this a bug or intended? More likely the latter case considering codex suddenly offered usage resets.

Same，five hours limited，can do a little work

Same issue, can't work more than 2-3 prompts in the weekly budget

Same issue

Yes this is definitely happening. I have just asked some questions (not related to my project, just questions like how to do this thing), and one for analysing my repo and have already consumed 25% of my 5 hour usage quota. I used GPT 5.5 medium for this without fast mode

+2, my 5x disappeared in the blink of an eye. Was switching to Codex a mistake? Today it burned through my quota even faster than the previous days.

+1, very very large difference in quota.

I saw this on reddit, and also found that my usage over the last couple of days seemed to exhaust my Plus plan much more quickly than expected. I thought it was because I was working on a different class of problem than usual and leaning more on high/gpt5.5 than usual.

I did some basic analysis - pulled sessions from ~/.codex in to duckdb. Results were inconclusive - cached vs non-cached vs output tokens (and not knowing how openai weight pricing behind the scenes on subscription models vs retail) meant that I didn't see anything similar to the findings at the top of the thread.

What I did notice, though, is that the session_meta.payload.base_instructions content (at least the one that OpenAI provides the client to use, not the one stored in the source code) has changed significantly. It's now much shorter. It changed for me some time between Jun 16 and Jun 21 - I didn't have any sessions between then.

There's nothing obvious about the prompts themselves, but it's interesting. I'm not going to reproduce them here as I'm not sure if that's against the TOS (as they're remote-provided, not in the source code). It's also likely that OpenAI may be A/B'ing these - so if this is an issue it's not going to affect everybody equally.

I use 5.4 High and same experience, horrible rate limits from last week

Same problem. 2 or 3 Plus accounts used to be enough. Upgraded to Pro plan and still draining very quickly.

╭────────────────────────────────────────────────────────────────────────────────╮
│  >_ OpenAI Codex (v0.141.0)                                                    │
│                                                                                │
│ Visit https://chatgpt.com/codex/settings/usage for up-to-date                  │
│ information on rate limits and credits                                         │
│                                                                                │
│  Model:                gpt-5.5 (reasoning medium, summaries auto)              │
│  Directory:            /*******/******/********/******          │
│  Permissions:          Workspace (Ask for approval)                            │
│  Agents.md:            AGENTS.md                                               │
│  Account:             *************@gmail.com (Plus)                            │
│  Collaboration mode:   Default                                                 │
│  Session:              019eefb9-5bd4-76e2-be90-b1f5****b51e1                    │
│                                                                                │
│  Context window:       50% left (135K used / 258K)                             │
│  5h limit:             [████████████░░░░░░░░] 58% left (resets 21:29)          │
│  Weekly limit:         [░░░░░░░░░░░░░░░░░░░░] 0% left (resets 23:48 on 24 Jun) │

╰────────────────────────────────────────────────────────────────────────────────╯

1 prompt ago, this was 100% of my 5h limit and 18% of my Weekly limit left.
My prompt was:

"In `website/views.py`, several dictionaries are used to build the landing pages.
Refactor them into a single page-indexed data structure. Since each page must have unique content, shared data such as `DEFAULT_PROCESS_STEPS` should be duplicated and adapted per page.
Also reorganize the first nesting level by template section, so each block matches the html sections where it is rendered.
Finally, update `ai_landing_page` to use the new structure.

At this point, I am not sure using Codex is improving productivity anymore. I am a worker and this is my job, I am self employed and can't afford to pay 200€+ plans, and being limited to one prompt every 5h and a total of 10 prompts a week makes Codex completely worthless and my investment in it a waste of money.
I feel like OpenAI is exploiting me and the other paying users, like a drug dealer they let us use it until we have been hooked on it, until we have it integrated in our daily workflows and then they just cut the limits off and wait for people to funnel more money into their system. I am canceling my subscription forever, and I will try Opencode and open source soultions.

+1 both codex and copilot pricing has gone to crap.

This system seems absurd, and it negatively impacts users. I changed my VPN country, and now my rate limit usage is lower. It worked fine in the other country without me changing anything, so it's likely my rate limit will be stolen again at some point in the future since the current VPN country might affect it again.

A more robust system is needed that doesn't rely on multiple servers to calculate usage. The current system is inefficient.

However, besides that, there must be something else affecting consumption because although weekly consumption has decreased, it still drops significantly for simple tasks like a 30-line file. This is in stark contrast to Claude, who only used 10% across 5 large tasks.

Why don’t they make an official announcement instead of doing this?

Why don’t they make an official announcement instead of doing this?

stop lowering your usage limits without notification!

10 or 20 fold increasing your costs of services over night is not a feasible business model anywhere.
If this was a bug I want my refund for additional credits, if this was intentional someones getting sued for false advertisement.
I also noticed my credits which I had to buy to complete a project in time (as a result of this) drained to zero without using any agent, I had checked before with some left, then tabbed back 20 minutes later and it was at 0..

This is absolutely wild.

Same for me. I was on the $100.00 5x plan and my project the next day suddenly started eating my 5 hour window at insane speeds I would say 10-20x as others mentioned. I had Hermes do some investigating and nothing conclusive was found I even moved to the $200.00 20x plan and it barely seemed to make a difference. I wish I had looked into it earlier and not assumed there was some problem with my workflow when I knew I had not changed much.

suddenly looks better to me, still usage last week was crazy, unusable, i already cancelled, but cant be real that there is not even a single reply from openai?

same here, i'm on $200 pro plan. I ran out of all the Tokens after just 2 days. Previously, I couldn't even use them for 7 days.

so sad @willwang-openai

Without a response, I'm still having problems.

Having the same issues... on Business Plan and 2 prompts use the whole 5-hour windows. The weekly limit was reached in three days, which is unacceptable.

Hey, same here. On the Plus plan, I normally generate 8 images at once, and it usually takes no more than around 30–40% at most. But recently, a single generation almost uses everything. On top of that, the images are coming out off-theme compared to what I defined in the MD file.

I’ve wasted a lot of tokens generating basically junk. Is it possible that something changed overnight? It honestly feels like yesterday it was working perfectly, and today Codex behaves like a completely different model.

It’s the exact same MD theme file — the only difference is the session (yesterday vs today), but the output looks like it’s from a different system entirely. I haven’t uploaded videos for several days, so this is hitting me pretty hard. The image quality is way more inconsistent now compared to before.

please check...

Definitely seems a bit better, it's actually doing some reasoning now

<img width="144" height="45" alt="Image" src="https://github.com/user-attachments/assets/792b301f-dc0f-4924-8b0b-175dc9f10c6c" />

I have no idea what's going on.
I'm testing on a fresh plus plan with gpt 5.5 high.
One prompt that read several text files, mostly markdown, several hundred lines. - used 100% of the 5h window and 16% from weekly limit.

Another test prompt to create "hello world" in python in an empty folder consumed 30% from a 5h window and like 7% from a weekly limit.

Codex Desktop 26.616.71553, Plus, gpt-5.5 high:

1) Sub-agent/profile task, 22 tools, 267k total tokens / 227k cached input:
5h primary: 1% → 99%, weekly: 0% → 16%.

2) “list folder files”, 1 tool (ls -la), 29k total tokens / 24k cached input:
5h primary: 1% → 9%, weekly: 16% → 17%.

3) “create hello script”, 3 tools, 64k total tokens / 49k cached input:
5h primary: 1% → 31%, weekly: 16% → 21%.
Biggest jump: +23% from one event with last_total=15,958, last_cached=15,232.

I'm pretty sure free cursor provides more usage.

Has anyone got an answer from OpenAI? Are you guys based in US or elsewhere?

Same pattern here, but using Hermes Agent. I think this might be on the server-side rather than the codex-client.

I noticed this as well, I could do a lot more work with codex on the plus plan before. Today I bit the bullet and got a Claude Pro subscription, and I can say that I definitely get more usage on Opus 4.8 Max than with GPT 5.5 xHigh. I always believed that OpenAI was the way to go if you wanted to get more usage out of your subscription, but this has been going on for too long without any form of public acknowledgment. If this doesn't get fixed in the next week before I get billed again, I am cancelling my subscription.

P.S. I believe this issue started after the banked resets feature was added to codex, but it's just a hunch and I don't know if it's related.

For those who’ve moved back to Claude because of this, what setup are you actually using day to day?

I’m used to Codex desktop with multiple project sessions. Are people mostly just running Claude Code in the terminal per repo, or using the desktop/web version to juggle several projects?

For those who’ve moved back to Claude because of this, what setup are you actually using day to day? I’m used to Codex desktop with multiple project sessions. Are people mostly just running Claude Code in the terminal per repo, or using the desktop/web version to juggle several projects?

I run both and you can use Claude in much the same way, just access it via the desktop app and with mobile access as well like Codex.

For those who’ve moved back to Claude because of this, what setup are you actually using day to day? I’m used to Codex desktop with multiple project sessions. Are people mostly just running Claude Code in the terminal per repo, or using the desktop/web version to juggle several projects?

The desktop app is terrible because I had to duplicate the skills, and it uses two routes for its skills, which breaks SSOT. People should work cleanly, following principles and philosophies, instead of randomly throwing things here and there. The CLI uses one route and can't find the skills used by the desktop app, nor does it read universal routes. I've already reported this, but it seems like they don't pay any attention to the issues.

At least I hope it reads the instructions from ~.agents/ although I don't know if it even reads that much or if we're still stuck with vendor only in a selfish way.

Codex is more flexible and reads the routes and instructions to the letter, uses the correct skills, and shows you the info when it does each thing.

I'm experiencing the same issue on Pro plan. 5h quota went from ~2% to 100% in minutes with only light usage.
Feedback uploaded — thread ID: 019efc9a-ac1c-7ea2-991f-837536660d06

I'm experiencing the same issue on Pro plan. 5h quota went from ~2% to 100% in minutes with only light usage. Feedback uploaded — thread ID: 019efc9a-ac1c-7ea2-991f-837536660d06

I have changed the VPN from Amsterdam to Spain, and now the weekly rate-limit drops more slowly although it is still bad compared to 1 week ago and the 5H thing is still bad

I had canceled Claude's plan and now I was forced to go back there and start configuring the AI ​​so that it doesn't hallucinate and create narratives in the jsdocs.

Same here, one simple prompt burned 50% of 5h limit on my Pro 20x sub...

Feedback ID:

019ef8d0-3aaf-7252-9b7f-5a1c3a6a9f3c

I'm experiencing the same issue on plus plan.

Same here, 3-4 Codex tasks, that usually are not complex, previously i would use maybe 40-50% of 5hr limit, and now i get to 0. and the weekly is not better.

补充一个「本机日志可提交」版本（Mac，UTC+8，本地时间）

我这台机器本地 ~/.codex/logs_2.sqlite（target=codex_analytics::client）里抓到了 7 条唯一 turn 的 gpt-5.5 token_usage 样本，且每条 turn_id 对应的日志记录被重复写了两次（同条目出现两次）：

| 本地时间 | turn_id | input | cached | non_cached | output | reasoning_output | total |
|---|---|---:|---:|---:|---:|---:|---:|
| 2026-06-24 16:46:08 | 019ef8cc-0ebb-7243-a1df-9ef33029ab82 | 519,681 | 427,904 | 91,777 | 10,642 | 4,617 | 530,323 |
| 2026-06-24 16:44:57 | 019ef8cc-2ce8-76d1-b792-dd9d804f0974 | 362,528 | 240,128 | 122,400 | 6,738 | 2,448 | 369,266 |
| 2026-06-23 23:19:51 | 019ef50e-a337-70e2-8fcb-fa0239a3504d | 398,906 | 321,280 | 77,626 | 7,885 | 3,619 | 406,791 |
| 2026-06-23 23:19:17 | 019ef50e-7ce8-7893-85ee-7cdb0eb27b1b | 473,639 | 402,688 | 70,951 | 8,043 | 1,876 | 481,682 |
| 2026-06-23 23:18:46 | 019ef50e-8ec7-77c0-854c-d74ab9ac364e | 336,395 | 267,520 | 68,875 | 5,131 | 2,310 | 341,526 |
| 2026-06-23 11:33:48 | 019ef287-6710-71c0-a5c0-d54f1cb42bad | 370,393 | 313,600 | 56,793 | 13,458 | 7,886 | 383,851 |
| 2026-06-23 11:19:15 | 019ef275-92d7-7062-a6e7-2c10cab23730 | 668,557 | 571,648 | 96,909 | 10,219 | 3,875 | 678,776 |

可见：

• 单条请求 total tokens 最高到 678,776
• reasoning 并不高（但 total 依然很高）
• 同条目都带 dropping analytics events: queue is full

这更像是“这批窗口里 gpt-5.5 的计量倍率或计量路径发生偏移”的客户端外显问题，而不是输入规模异常。建议这部分也并行排查：

• 5.5 计量是否按历史公式变动了
• 是否有重复计量（turn 重复日志是否代表重复发送/重复计费）
• 为什么日志里出现 queue full，是否与消费风暴同步出现

如果需要我可以再导出包含 feedback_file 的原始行（完整 1:1 原文）供你们做比对。

I barely do anything on it. I always start fresh sessions and yet it swallows great amounts of tokens.

I have never hit my quota so far but this week I hit my quota in just 4.5 days, my usage pattern is the same. Something is wrong and OpenAI is definitely nerfing all the plans and usage.

same here, around 10x to 30x quota usage increase. on top of that, the compaction is awful and model overwrites files it already touched in past and completed.

I agree. I believe openai has realized what a steal the Plus plan used to be, and they have now started to rate limit it to lower computing costs. It's still not as bad as Claude however.

Adding another data point from a Pro account. I inspected local Codex rollout-*.jsonl logs and compared token_count / rate_limits events against prior months.

Environment:

• Plan: pro in rate_limits.plan_type
• Surface: Codex Desktop / VS Code integration
• CLI/app log version seen in session metadata: 0.142.0
• Local timezone: Europe/Bucharest / EEST

What looks wrong:

• Historically I did not hit these limits with similar or heavier usage.
• June is elevated, but it is not my highest local token-usage month.
• The limit pain started feeling much worse despite usage not being uniquely high compared with earlier months.

Month comparison from local token_count.last_token_usage logs:

| Month | Raw last_total sum | Approx uncached input + output | Active days |
|---|---:|---:|---:|
| 2026-03 | 28.0B | 1.53B | 24 |
| 2026-04 | 9.47B | 706.8M | 23 |
| 2026-02 | 6.31B | 243.6M | 16 |
| 2026-06 | 5.34B | 288.1M | 22 |
| 2026-05 | 3.78B | 151.6M | 20 |

So June is not low usage, but it is not an outlier versus months where I did not experience the same quota pain.

Daily comparison:

• 2026-06-03: ~597.8M raw last_total
• 2026-06-22: ~582.9M raw last_total

Those are comparable local-log usage days, but June 22 was the day I had to use the new OpenAI reset/banked reset function because both weekly and 5-hour limits were effectively unusable.

Reset/accounting anomaly:

• Before manual reset on 2026-06-22 17:48 EEST: primary.used_percent=39, secondary.used_percent=96, old weekly resets_at.
• Immediately after reset on 2026-06-22 17:49 EEST: primary.used_percent=0, secondary.used_percent=0, new weekly resets_at.
• After that, active sessions showed inconsistent/stale state for a while: some token_count events reported the new reset counters, while other active events still reported old or exhausted weekly state near 96-100%.
• Later the stable new weekly window showed secondary.used_percent low, but the 5-hour primary window was still climbing much faster than expected for my historical workflow.

Current account state from a fresh check:

• A new thread started at primary.used_percent=20, secondary.used_percent=80.
• During that thread, local total_token_usage.total_tokens increased from ~31k to ~536k while the displayed counters stayed 20% / 80%.
• This makes it hard to reason from local logs alone whether the problem is quota weighting, stale reset state, duplicated accounting, or background/subagent accounting.

My read:
This looks related to the reports here and in #28823, but with an additional reset-state angle. The manual reset did reset the visible counters, but local logs show mixed old/new rate-limit state across active sessions afterward. Also, my June usage is not uniquely high compared to earlier months where I did not hit limits like this.

What would help:

• A per-turn ledger showing model, surface, bucket charged, input/cached/output/reasoning tokens, tool/subagent/background charges, and any multipliers.
• Clarification whether banked/manual resets are expected to immediately invalidate stale quota state across all active Codex sessions.
• Clarification whether rate-limit weighting changed around June 16-18 or with the reset/banked-reset rollout.

I agree. I believe openai has realized what a steal the Plus plan used to be, and they have now started to rate limit it to lower computing costs. It's still not as bad as Claude however.

In my case, Claude now consumes less 🤣

nah claude is much better since opus 4.8, only 4.7 was bugged tokenizer and inefficient

literally everyone

nah claude is much better since opus 4.8, only 4.7 was bugged tokenizer and inefficient

can you use the MAX with Hermes agent like Codex?

Same for me. I'm at 3 Plus accounts and they all drain faster than ever before. Something is off.

I've also noticed my usage limits draining HEAVILY, I'm on a Pro 5x Plan.

It seems to depend on region and/or time of day

I upgraded from using 2 Pro accounts to 4 Pro accounts because I was hitting the limits too often. Surprisingly, even with 4 Pro accounts, I’m reaching the rate limits much faster than I did when I only had 2. The token usage seems to be draining significantly faster than before. Something definitely feels off.

I upgraded from using 2 Pro accounts to 4 Pro accounts because I was hitting the limits too often. Surprisingly, even with 4 Pro accounts, I’m reaching the rate limits much faster than I did when I only had 2. The token usage seems to be draining significantly faster than before. Something definitely feels off.

Jokes and complaints aside, what are you building with 4 Pro accounts mate? 😅

I'm on the 20x plan, and the last couple of months I was able to get about ~5 days of usage out of it each week before running out of credits (I push it pretty hard). Last week I ran out of tokens day 1, which seemed strange, but I thought maybe it was a new set of skills I was trying. Uninstalled them, went back to basic (no skills/plugins) for this week. Ran out of tokens day 1 again.

I've cancelled the sub, since I'm not paying $200/mo for 1 day of usage.

Adding a data point: Pro Max ($200/month) subscriber. My weekly Codex quota was fully drained within one week with minimal actual usage, consistent with the overcounting described here. I renewed my subscription but the weekly quota was not reset after renewal — still showing 100% used. Submitted a support ticket, awaiting response.

All I'm saying is...if you haven't switched by now...you missing out.

Hi all - VB from OpenAI here, can you please share your /feedback IDs or User IDs, that'd really help us investigate better 🙏

Thanks, VB. My feedback ID is: 019eef1b-3eca-7c00-9202-22887185aa6c

X/Twitter handle: https://x.com/CalimanuLoredan

Dom from OpenAI also reached out to me over X DM, and I already shared the main details there. The affected session I pointed to is:

019eef1b-3eca-7c00-9202-22887185aa6c

The strongest anomaly I flagged was around 2026-06-23 10:13:46 -> 10:14:08 EEST, where the 5h meter jumped from 35% to 76% and weekly from 13% to 20%, while local last_token_usage.total_tokens on the latter event was about 134,059.

Also relevant: after using the manual/banked reset on 2026-06-22 around 17:49 EEST, local logs showed mixed old/new quota state across active events.

same for me. I just realized how often token usage limit warning is popping up in these days

same for me. pro 5x weekly gone just like nothing, 5.5 low

Same ###

I recently built a full OT asset pipeline and worked on it intensively for a full week without hitting the limit even once. Large prompts, code changes, analysis, refactoring, debugging everything worked fine. +-10.000 lines of code

Now it feels completely different. If I ask just two normal questions, for example something simple like “can you add JSON to CSV or XML output,” I almost immediately hit the 5-hour usage limit. At the same time, I already seem to be around 20% of my weekly limit.

That does not feel normal compared to how it worked before. It does not look like I am suddenly “using more”; it feels like the same type of prompts are now being weighted much more heavily, or that something changed behind the scenes in how usage is calculated.

For me the difference is very clear: before, I could work on a complex pipeline without issues. Now, even a small change request consumes a huge part of the limit almost immediately.

Here are my feedbackID's for both high token consumption and terrible instruction following /overall dumbness following.

Feedback ID (high usage consumption for a shorter running task compared to what I was running a few weeks ago):
019f02b5-68b6-7073-b3aa-023c9465d05a

Feedback ID (model seems unconfident in it's answers, constantly switching its stance on a topic, indicating that it was never really grounded in a firm conclusion in the first place):
019ef1e6-9139-72f3-9b8e-f8c5d81b8f62

FeedbackID (model does not folllow instructions and seems to do thing extremely surface level requiring a lot of hand holding):
019ef476-4e00-7693-820c-04e4fd6fd7e8

(Model quality degradation and excessive usage compound when running /goal . Overall the implementation becomes really messy and uses up all your usage limtis):
019eed37-a313-7982-81da-3697c058bc91

I'm on the pro 5x plan and the usage is just so unnecessarily high. Furthermore, the instruction following has gotten noticeably worse since around 16th of June. What the model could have previously done easily in one shot now takes multiple turns to complete. It's almost as if the model is starting to exhibit laziness in it's responses, refusing to double check and verify like it normally does before providing it's outputs. It's planning and implementation has also become a lot more surface level. I understand that, as the codebase grows, the model may find it harder to complete it's tasks. However, the problem here is that even relatively easy tasks are now taking many more turns to complete, and a lot more hand holding. This issue then compounds with the fact that each turn now consumes such as drastically high level of tokens. As a long time codex user, this is the first time I'm experiencing such severe degradations in the user experience. The last week has been utterly terrible, and not worth the price of a pro subscription at all.

Same problem. Feedback id:019f03d4-fd55-7442-8e3a-eb2c2e133a47

Same issue here — this is not isolated to a single Codex session.

I experienced abnormal Codex quota / 5-hour limit drain across multiple sessions. Here are representative /feedback IDs from affected sessions:

019f0120-735c-7a92-ac70-324c946fe315
019efa24-d8d2-7ad0-9824-5d60a3f1d778
019f00ac-86cf-77f0-8217-751a6ea40e7d
019ef4ff-83ee-76f2-b3b3-4e2586027d08

Adding a Pro 20x / $200 account data point.

OpenAI/Codex account ID: 63fe18c3-62b9-4801-a8c3-fb97a28d9860

I do not currently see separate /feedback submission IDs in my local logs, but the affected Codex session/thread IDs are:

• 019efdaa-6daa-7bd3-b0c4-eb976eced4a1
• 019ef3a5-7375-7522-8f6f-c8458d9f86d3
• 019efd2e-85e4-7c52-bf22-c0339d6611f5

My main issue is not failed requests or retries being charged. These were normal successful task runs, but the quota drained extremely quickly.

On 2026-06-25 afternoon, roughly 12:00-18:30 Asia/Shanghai, my local Codex/CC Switch logs show:

• 709 Codex requests
• all 709 returned status_code = 200
• input tokens: 92,862,139
• output tokens: 375,205
• cache read tokens: 86,210,432
• total local token footprint: 179,447,776
• 291 requests were above 300k token footprint
• two sessions alone accounted for 117,053,830 token footprint, about 65.2% of the afternoon total

The two biggest affected sessions were:

• 019efdaa-6daa-7bd3-b0c4-eb976eced4a1: 2026-06-25 15:25:42-16:30:59 CST, 216 successful requests, 60,627,004 token footprint.
• 019ef3a5-7375-7522-8f6f-c8458d9f86d3: 2026-06-25 14:25:08-15:22:26 CST, 161 successful requests, 56,426,826 token footprint.

Locally these sessions were gpt-5.5 with xhigh reasoning. From the product UI side, my weekly quota appeared to drop by about 30% in a single afternoon on a $200/Pro 20x account.

This feels consistent with the reports here: normal successful usage is being weighted much more heavily than before, or the quota accounting is overestimating usage. Please investigate the above account/session IDs.

Feedback id: no-active-thread-019f0430-d1f5-7d72-b684-6bd3f8146b33

Pro 20x account here 58% of weekly used in 1 1/2 days, not even using fast.

Same here:

Feedback ID: 019f047a-33a1-7ce2-b0b0-d8200f2291d3

I thought I was going crazy, but in the first session after my last weekly reset the 5h limit and weekly limit were dropping at about the same rate.

Looks like (for me at least) it's related to the vscode codex extension. When I use codex through the vscode extension, my usage is consumed super fast. I swapped to OpenCode after a reset and now my usage is barely dropping

@cjmanca did you try codex desktop app?

I can confirm Codex desktop app has this issue.

Hahaha!

"We have identified that some reports of Codex usage limits depleting faster than expected _are related to our abuse and fraud prevention systems incorrectly rate limiting certain accounts_. The impact appears limited, and we have not observed broader degradation to Codex usage. We are continuing to monitor the situation."

https://status.openai.com/incidents/6enf4645

Well that's not going to help. 🤦

i can confirm

I am on the Pro plan and experiencing a serious billing/syncing bug today.

Here are the facts regarding my usage sync:

• My first 5-hour session fully used up its limit, and my weekly limit dropped from 100% to 70% (a 30% drop).
• During my next 5-hour session, I still had 54% of my weekly limit remaining, and my 5-hour limit was sitting at 12%.
• Then, right before my eyes, that 12% instantly vanished to 0%, causing a total lockout.

The 5-hour bucket suddenly glitching to 0% is definitely a server-side bug.

This is absurd tbh, I have two separate accounts that suffered the same issue these past few weeks, and I even bought x2 $100 credit, which depleted in less than a day while working on a project.

i used 8% of my 5h limit and 7% of my weekly limit somehow. in one prompt, pro subscriber

Lost 13% of 5h and 2% of weekly in less than an hour after reset by just having Codex open, no prompts.

I just got a free rate limit reset. It says when I run out I can reset it and keep working. Has any body else received this?

Yeah, Tibo announced it last night / this morning (for values of 'morning' 😉 )
(That's how 'resets' have been implemented now for a couple of weeks. No doubt, in part, because of the hoo-ha on the likes of https://github.com/openai/codex/issues/9508)

https://x.com/thsottiaux/status/2070653282440405046?s=20

<img width="445" height="317" alt="Image" src="https://github.com/user-attachments/assets/166b0200-31a8-4769-93f9-3d8f5973c752" />

It's great that they gave us free resets but I do hope they will fix the main bug/issue as well...

The novelty - and randomness - of the resets wears off real quick. 😆 👍

To add another detail, I only started noticing the severe increase in usage a couple of days ago, maybe because I'm in Europe the bug/update reached me later(?). Previously I never hit a usage limit since I upgraded to the pro x5 plan. I just noticed the attached image in analytics, a lot of use of gpt-5.4 and I have never selected the 5.4 model.

Also note the peak from June 21, I didn't hit the 5h limit or the weekly limit. My weekly limit was reset on the night of June 25, on June 26 I hit the 5h limit for the first time, and I'm currently at 53% of my weekly usage limit at the time of writing (June 27). This is insane.

<img width="850" height="450" alt="Image" src="https://github.com/user-attachments/assets/062b4065-7610-4b36-8c29-bf83ea018821" />

<img width="850" height="408" alt="Image" src="https://github.com/user-attachments/assets/06eae3a3-9326-4d20-ba8c-cc4b694d21c1" />

I launched the codex app and didn't even touch it - I literally stepped away from my machine - and when I got back after ~2 hrs I noticed my quota had drained to 10%.

I started trying to figure out what happened, and while I was doing that the remaining 10% drained to 0% in about 10 minutes.

I have 0 recurring jobs or anything like that. The dashboard claims it was all "Desktop App" usage, but you can see from my session logs I haven't touched Codex in days.

<img width="1125" height="988" alt="Image" src="https://github.com/user-attachments/assets/4ca6c7c5-04e9-47a8-acf9-03a2f6f3a3c3" />

I've noticed usage being depleted slowly "without usage" too :)

Yes. Literally just having the app open drains usage.

My quota is back to stable now. Looks like they are rolling out patches

I agree it does seem that OpenAI has fixed the issue.

Something is definitely wrong. Not only did I notice some usage when doing nothing, just generally my usage is being eaten at a much higher rate — like, maybe 30 minutes of work, where I used to go for a couple hours at least before hitting my 5h limit. It's like usage wasn't just cut in half, I could stomach that even though it sucks. Now it's like, 4-5X usage? Is this really the new reality?

Something is definitely wrong. Not only did I notice some usage when doing nothing, just generally my usage is being eaten at a much higher rate — like, maybe 30 minutes of work, where I used to go for a couple hours at least before hitting my 5h limit. It's like usage wasn't just cut in half, I could stomach that even though it sucks. Now it's like, 4-5X usage? Is this really the new reality?

They have claimed that it's being fixed and they are rolling out free rate resets. I think different places are receiving different times of the bugs and fixes

Any official announcement from OpenAI? I can't find a reliable source that documents this is actually a bug -- seems strange given how many users this affects...

Any official announcement from OpenAI? I can't find a reliable source that documents this is actually a bug -- seems strange given how many users this affects...

Yea here is their official link: https://status.openai.com/

> Any official announcement from OpenAI? I can't find a reliable source that documents this is actually a bug -- seems strange given how many users this affects... Yea here is their official link: https://status.openai.com/

Ongoing for 2 days? It's happening since 16th of June.. 13 days..

"The impact appears to be limited"

Yeah to thousands of users... at least acknowledge that this is a serious issue.

much more than thousands of users lmao

+1, same pattern here on Pro (100 EUR/mo): 5h meter jumped instantly from 80%+ to 0% in 1-2 messages, no warning. Also tracked in #29895 and #31125, and cross-referenced against status.openai.com incident 01KW2E6W0503W4NXJNCVAG8V6T.

i had to move on Claude, same pattern, high cost for same prompt that I send everyday. WT* are happening?

i had to move on Claude, same pattern, high cost for same prompt that I send everyday. WT* are happening?

The squeeze?

Windows 11
Codex Plus Plan
GPT 5.5 Low

I just wrote the word "Hello" and left the PC. Codex has used up 20% of the 5 hour limit and 5% of the weekly limit. For saying hello to him.
No MCPs are enabled. There are no agents and other things. I'm using Antigravity now,
GOODBYE Codex

Same here I am using plus plan and it is evaporating during discussion session after 10 questions

Is it just me, or does Codex still feel like it's using a lot more of my quota compared to a month ago. Yes, it's better than a two weeks ago when we had a massive problem, but it still drains my quota faster nowadays.

100%.

I have a workflow I've been using for months.

Previously I could chat with codex for ~1 hour before being rate limited.

Now I'm lucky to get 6 questions before using my 5hr quota, using the exact same prompts and skills.

I still have the sessions in codex from before this started showing this huge disparity.

Currently browsing OpenRouter for alternatives, this is ridiculous.

today the 5 hour limit is gone and since just a little bit ago limits seem to be ok for me again (im on 99% now for quite some time while yesterday some simple prompts ate a lot of pro lite....

The 5-hour limit removal is currently touted as temporary. With the ChatGPT5.6 rollout, Codex App -> ChatGPT App migration, and announcement of 5.4 retirement... it's been an absolute roller-coaster of quota resets, random performance, and your-guess-is-as-good-as-anyone regarding results weekend.

This is blatant "move fast and break things" strategy (with a bit of Claude vs Codex PR thrown into the mix). Don't bother testing, or releasing a coherent - or even stable - product; just shite it out the door any time it suits Marketing - and have subscribers PAY to do the testing and work through the pain points. //slow-clap

Same Weekly Codex tokens gone in two days with a pro account.

I'm burning through my weekly limit (that used to last a week) in a day (50% of my weekly usage in ~2 hours so far today). I'm downgrading myself to Terra for now and hoping this gets fixed, otherwise I'll need to supplement with Claude.
My 'turns per week' has dropped from ~300 previously to ~70.