Session limits draining significantly faster

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

• #23089
• #23068
• #24336
• #23136
• #23297

Powered by Codex Action

Exactly the same here : feedback ID : 019e5a8d-7d8e-7d73-bd64-c5f31a41c208

Codex IDE Extension regression after recent Codex rate-limit incident / weekly reset.

I am on ChatGPT Pro, using Codex IDE Extension in Cursor.

This is not only about Speed/Fast mode being more expensive. I was already using Speed before, with the same kind of workflow. The regression is that, compared with my previous normal Speed usage:

• Codex is now significantly slower.
• Credits/tokens are draining much faster than before.
• The experience is worse even though I did not intentionally change my usage pattern.
• This is worse after my weekly usage reset.

Observed symptoms:

1. Abnormally high usage drain

• My weekly tokens/credits were reset.
• After the reset, usage is still depleting extremely fast.
• It appears to drain faster than before the reported Codex rate-limit incident.
• Please verify whether my Pro usage bucket, weekly reset, and Codex promotional/included limits are correctly applied.

1. Speed/Fast mode regression

• I was already using Speed before.
• Previously, Speed felt faster and the usage drain was manageable.
• Now, even with Speed, the IDE extension feels slower while consuming more credits/tokens.
• Please check whether Fast/Speed mode accounting or routing changed, or whether my account is affected by abnormal billing/usage accounting.

1. Missing context UI !

• The previous context icon/control disappeared from Codex IDE Extension in Cursor.
• I can no longer see the same context indicator/control as before.
• Please confirm whether this is an intentional UI change or a regression.

1. Slash commands not available

• In the Codex IDE Extension composer, typing /auto shows “No commands”.
• This seems inconsistent with the Codex IDE Extension documentation, where slash commands such as /status, /feedback, /local, /cloud, and /auto-context are expected.
• Please check whether the slash command registry is failing to load in Cursor.

Environment:

• ChatGPT Pro
• Codex IDE Extension
• Cursor IDE
• Windows + WSL2 workflow
• Local mode: Work locally
• Model shown in UI: GPT-5.5 High
• Speed option visible
• Screenshot available: composer shows /auto and “No commands”

Please investigate:

• Whether my Pro Codex limits and weekly reset are correctly applied.
• Whether the recent Codex rate-limit incident mitigation fully applied to my account.
• Whether Speed/Fast mode is over-consuming credits/tokens.
• Why Speed is now slower than before while consuming more.
• Whether the missing context icon is expected or a UI regression.
• Why slash commands show “No commands” in the IDE extension.

<img width="656" height="186" alt="Image" src="https://github.com/user-attachments/assets/89409f58-c87f-4a5b-9175-0728fe3d5352" />

<img width="215" height="254" alt="Image" src="https://github.com/user-attachments/assets/5e7c97c3-5eea-43bf-8199-21a0b15fd6da" />

Adding more context after some testing:
Even in a fresh thread, 5% drain in 2.5–3 minutes of basic work (file reading, research, simple instructions) is already more than it should be. But doing similar work in a long running thread hit 13% in roughly the same time on the same model and effort setting. That's a big gap for essentially the same task.
The fresh vs long thread difference is what makes me think caching is the likely culprit here. If caching was working correctly, longer threads should get cheaper over time as the context builds up and gets cached, not more expensive. The fact that it's the opposite points to something being off there.
@MedEqz is seeing the same thing independently, which is good to know, not just a one account thing. And beyond GitHub, there are a lot of people on X reporting the same issue of limits draining way faster than expected, and it's still happening even after the May 23 incident was marked as resolved. So this doesn't seem like an isolated case at all.

Screenshot: 13% of the session limit used in a 2-minute 43-second task of simple Google searches and some local file reading
<img width="1488" height="504" alt="Image" src="https://github.com/user-attachments/assets/e4e36a08-76d7-4b40-8b7c-79466b1812be" />

Same issue here, I even tried 5.4 Low and it still took like 18% of 5h usage limit after running for 2 min. I'n on plus plan.

cry about it, openai gotta make the bands

@nika34-glitch

Switched to Claude Code, much better limits there right now. Saw it coming, so I didn’t waste much time switching.
According to my personal calculation, Claude is giving 50% more inference than Codex on the 5x right now, they'll be roughly the same after the 50% promo ends in Claude code though.