Paid Usage Still Dropping too Quickly
What issue are you seeing?
My weekly usage completed yesterday [2026-03-14] around 1:00pm where I purchased an additional $40 of credits. I burned 80% of my credits doing a small portion of work [1000 Credits > 208 Credits].
Overnight my OpenClaw system ran two GPT 5.1 Codex Mini prompts and one GPT 5.4 prompts dropping my credits from 208 to 178. This morning after three prompts to my OpenClaw System [GPT 5.4] my credits dropped from 178 to 161 going over the "hover" effects 4 credits per prompt max [161 credits left hover showed 40 local prompts available worst case scenario].
If you take a look at my usage this clearly indicates as problem either with the graph representing usage, or the amount of credits that are being used.
<img width="1674" height="1887" alt="Image" src="https://github.com/user-attachments/assets/3daf2b28-98da-4cc5-8c73-9c67e519ee15" />
The bar for 2026-03-14 [Second from the right] indicates I used a typical amount of usage as a normal day before the weird spikes from issue #13568, and the subsequent weekly resets. My Plus plan will usually last me four days a week and one purchase of the $40 credit will cover the deficit for a month and a week. If we take into account the 30% faster usage for using GPT 5.4 the $36 of credits used should last me substantially longer than one day at typical use.
While all my usage was through OpenClaw on 2026-03-14 and 2026-03-15 the graph clearly indicates I did not do a substantial amount of work compared to my typical days.
What steps can reproduce the bug?
- Setup OpenClaw locally on your computer.
- Attach OpenAI-Codex O-Auth to your OpenClaw installation.
- Set GPT 5.4 as your default model.
- Use up all weekly allotment of Codex Credits on a Plus account.
- Purchase an additional $40 of top up to get 1k Credits.
- Use the following token amounts per model [might be more as this doesn't include chatting tokens]:
• gpt-5.4 → input 2,018,178 | output 188,938
• gpt-5.1-codex-mini → input 324,003 | output 38,983
- Verify this brings credits from 1000 to 208
What is the expected behavior?
Assuming standard API pricing, no cached input, and not Batch/Flex/Priority just using the API would cost:
gpt-5.4 at $2.50 / 1M input and $15.00 / 1M output:
2,018,178 input = $5.05
188,938 output = $2.83
Subtotal: $7.88
gpt-5.1-codex-mini at $0.25 / 1M input and $2.00 / 1M output:
324,003 input = $0.08
38,983 output = $0.08
Subtotal: $0.16
Total: $8.04
Which would be about 20% of my $40 credit assuming calling the API is the same as using Codex's credits.
Lets take the worse case scenario, I used twice the amount of GPT 5.4 tokens, as the above doesn't take into account chat usage, and applying the "Long Context Single Session Surcharge" [Note: I used new sessions often]:
gpt-5.4 Doubled usage: 4,036,356 input and 377,876 output
Base pricing is $2.50 / 1M input and $15.00 / 1M output, and for GPT-5.4 sessions with more than 272K input tokens, OpenAI charges 2x input and 1.5x output for the full session. That makes the effective rates $5.00 / 1M input and $22.50 / 1M output.
Input cost: $20.18178
Output cost: $8.50221
Subtotal: $28.68399 ≈ $28.68
gpt-5.1-codex-mini
Doubled usage: 648,006 input and 77,966 output
Pricing is $0.25 / 1M input and $2.00 / 1M output.
Input cost: $0.1620015
Output cost: $0.155932
Subtotal: $0.3179335 ≈ $0.32
Total: $29.0019235 ≈ $29.00.
This would be about 72.5% of my $40 credit assuming calling the API is the same as using Codex's credits. While this is much closer to the amount used this usage does not reflect in the usage graph.
Additional information
I'm currently speaking to support via email with Case Number: 06791252.
I might be delayed responding here. If you'd like me to test anything or provide additional information feel free to email me at the address attached in the Support Ticket.
2026-03-15 19:16 Edit:
I don't believe this issue is a duplicate of #13854 as I have not been using Spark/changing models in Codex and this issue is not using the Codex App.
I've been attempting to use GPT 5.4 in OpenClaw and Codex to see if this usage drain is more of a graphing issue or probable backend issue.
<img width="1268" height="1611" alt="Image" src="https://github.com/user-attachments/assets/a1cb101c-4577-4643-8f9b-63eba52615f0" />
I've used OpenClaw and Codex for an hour today and it almost used the last 20% of my credits. Looking at the graph it looks to be ~25% the height of the previous days usage.
I don't believe this is a graphing issue. There's definitely something wrong where either $40 of credits provides substantially less than half the weekly usage of the $20 Plus plan or usage is being consumed at an alarming rate.
Side Note:
Unsure if related - I've noticed a dramatic decrease in GPT 5.4's intelligence today. Substantial hallucinations when asked to launch Cron jobs [OpenClaw], "Laziness" in terms of following my instructions and implementation [Codex].
2026-03-16 Edit:
I realized the usage dashboard clearly shows the issue:
<img width="1278" height="1605" alt="Image" src="https://github.com/user-attachments/assets/8eaf1aa6-4852-492c-b56c-e2730c28d315" />
<img width="1007" height="579" alt="Image" src="https://github.com/user-attachments/assets/d40e4615-2bb7-4cbb-89bf-d4fe359ce473" />
I purchased 1k in credits on 2026-03-13 12:01PM and in the usage table I've only used 534.89. That doesn't take into account I used Codex substantially before 12:01PM that day.
``4.11 + 9.825 + 125.905 + 74.505 + 320.545 = 534.89``
20 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
<img width="1030" height="362" alt="Image" src="https://github.com/user-attachments/assets/11dccaa9-a97a-4f93-aa8a-525e4bf64f6b" />
The last day on chart, May 15th, is
619,79 creditsspent on 5.4, about30% medium, the rest ishigh. Confirms the statement.<img width="695" height="215" alt="Image" src="https://github.com/user-attachments/assets/56450813-430e-4ee3-b405-d050a18534bf" />
me too, May 15th usage drop from 47% to 3%, but May15th I use very little. I also use gpt-5.4 on the Mac Codex app, pls check it.
<img width="2184" height="751" alt="Image" src="https://github.com/user-attachments/assets/10b906c0-9f26-4615-9d1f-34f5a2431557" />
Before the 2x I wouldn't have to worry about blowing my weekly. For two weeks during the 2x period I wouldn't even drop below 50%. For the past roughly 1-2 weeks I've been having to watch my usage very closely as even with Codex 5.3, usage is dropping much faster than I'm used to. One prompt the other day with 5.4 dropped the usage by around 5-10%, that shocked me.
@etraut-openai
I refer here to the closed PR #13568: in my case there were no experimental features enabled, I don't use
/reviewand fast-mode.If that's +30% consumption, then maybe you should state it clear in UI.
Otherwise, it looks like OAI engaged people that there is an issue on their side, and now the only people who knows it's greedy model are those who read github issues after they hit the wall.
@iw2rmb they stated it via: 5.4. eats 30% more, sub-agents eat more and thats where its probably coming from.
I am a bit confused since the claim was that 5.4. is supposed to be more token efficient, but apparently not on
xhighandhigh.For the first time this week I've hit limits 2 days before reset after using Codex (mostly on xhigh) since 5 months full-time.
Something is definitely wrong but I think its currently justified with their communication and this thread: https://github.com/openai/codex/issues/13568
<img width="2574" height="942" alt="Image" src="https://github.com/user-attachments/assets/5fb40353-330b-40a9-a574-7d34035d1d76" />
Usage was kind of the same. Only 1 month ago it was very hard to hit any limits whatsoever but now its a struggle even with the claimed 2X.
User:
user-EwwvXRbDXkpKEnwW9TeVbKGT<img width="1225" height="364" alt="Image" src="https://github.com/user-attachments/assets/e2b8dfb2-f673-498c-a9c7-5e5a6a7f0b67" /> Note: Today’s usage (March 16) only appears normal because it is not even the middle of the workday yet.
Something has been off since yesterday. I’ve been working all week at the same intensity, usually in a single thread about 90% of the time. I’m using 5.4 xhigh in the desktop app, with standard context length, no MPCs, literally the way the app is shipped through the Microsoft Store.
Everything was fine since last reset on previous days. I was using fast xhigh all day, in a single thread, with no issues and normal usage. But yesterday it suddenly started burning through multiple percentage points within minutes. Now I’m on standard mode instead of fast, and it’s not even midday, yet I’ve already reached what used to be my average full-day usage when running fast the entire day. I am just mid week, and if this continues by tomorrow I will be out of usage limits on PRO plan even on standard (mind 41% still left, but in the last 24 hours or so it went 70%+ to 40%.
I'm starting to think this is all intentional to adapt users to the new limits when the promo is over.
Having the same exact issue. On 3/14/26, I sent 3-4 prompts, and I had already hit my limit. Can't even blame the whole "2x ending" thing because look at my prior usage... I came no where close to half that.
<img width="1067" height="421" alt="Image" src="https://github.com/user-attachments/assets/ccb5fec6-21fc-4127-8c3f-1af00d6e89c9" />
I think the promo was not even working for a certain % of users for like 3/4 of the period :)
Also another thread: https://github.com/openai/codex/issues/14815
Running /review a single times takes 50-100 credits, so if you use that, they go away quite fast.
Same issue on Pro.
<img width="1886" height="1763" alt="Image" src="https://github.com/user-attachments/assets/ba0f85df-0635-44cd-bc66-dfc4f28ca449" />
i bought pro and it feels that i have less tokens than 2 plus accounts.
I literally have 3 pro accounts and two plus accounts, and 2 plus accounts give me more token usage than the pro accounts. I don't even know how this makes any sense. This needs to be addressed ASAP.
<img width="1272" height="1195" alt="Image" src="https://github.com/user-attachments/assets/7b1cbdd4-f3f9-437a-a13b-434ee3adbc86" />
Here is another Pro account
My 5-hour usage limit and weekly usage limit are syncing with a 1% difference.
That's strange.
It feels like they're changing limits weekly. Last week was actually good, I could get work done and not worry about quota but the past two days alone I've used 30% of my quota and I've barely used Codex, literally all I've had it do is update an unsloth training script a few times. "2x" limits my buns, they're def changing limits when they feel like it
@SJY051 & @gxvingates are you using GPT-5.4 when you saw the extreme usage? It looked like that for me too with 5.4 and then switched to 5.2 and felt normal again...
@AlexisDelatolas Yes I mainly used GPT-5.4 xhigh, with no issues for a while, then was shocked by my weekly limit plummeting after a reset a few weeks ago and had to start using Codex-5.3 high/medium. This week I've been using the hell out of 5.4 xhigh and I have 36% remaining, no complaints this week. I understand xhigh is going to consume a lot, but when it varies THAT greatly between resets, it's just a bit confusing
Great to know that 5.4 is decent again! I might give it another try when my week gets renewed...
We're using #14593 to track this issue. I'll post there when we have updates to share.