# Codex Rate Limits Discussion Thread

**URL:** <https://community.openai.com/t/codex-rate-limits-discussion-thread/1378553>\
**Category:** Codex\
**Tags:** rate-limit\
**Created:** [April 4, 2026, 10:54pm UTC](https://community.openai.com/t/codex-rate-limits-discussion-thread/1378553 "2026-04-04T22:54:52Z")\
**Posts on this page:** 1\
**Showing post:** 504

<div class="post-metadata">

**Author:** ![krasnik](https://avatars.discourse-cdn.com/v4/letter/k/71e660/32.png) [@krasnik](https://community.openai.com/u/krasnik)\
**Post date:** [August 25, 2026, 2:35pm UTC](https://community.openai.com/t/codex-rate-limits-discussion-thread/1378553/504 "2026-08-25T14:35:05Z")

</div>

Confirmed live on Plus now. This isn’t just an announcement anymore — my Usage page now shows the restored **5-hour limit** , alongside the weekly limit, shared across Codex, Work, Workspace Agents and ChatGPT for Excel.

 ![image](https://us1.discourse-cdn.com/openai1/original/4X/c/f/2/cf26e178280ccc551d0d4823407bef5d7d0e53a2.png)

This creates a **particularly bad** situation for long-running work.

I’m separately investigating an ordinary-Chat execution regression where my preserved historical GPT-5.6 Chat had worker-routed execution lasting **102m26s** , while current genuine GPT-5.6 Thinking turns repeatedly land around **1541–1564 seconds (~25–26m)** with work unfinished.

Work still receives the durable worker execution path on the same Plus account. But now **that** route is back behind the 5-hour gate.

So for heavy Plus workflows the practical choice is becoming:

ordinary Chat → long tasks currently hit the ~26m foreground execution class  
Work/Codex → proper worker execution, but now the 5h gate is back

Separate issue, separate evidence, but together they make genuinely long autonomous work much harder on Plus.

**I’d also really like OpenAI to clarify the current active-task behavior at the 5h boundary. Codex previously had a grace behavior where an already-running task could continue beyond the nominal limit. Current user reports indicate quota exhaustion can now strand active work instead.**

If the 5h limit is needed for compute smoothing, please at least make it optional against the weekly allowance, or guarantee that an already-running task can reach a clean stopping point.

My ordinary-Chat execution evidence is here:

> [@GPT-5.6 execution-routing regression: ordinary Chat stays foreground (~26m) while Work gets Temporal/SAServer worker handoff on the same Plus account](https://community.openai.com/t/1391943):
>
> Reproducible ChatGPT Plus execution-class regression: historical ordinary Chat used durable worker routing (\>100m); current ordinary Chat is non-Temporal/foreground (~26m class), while Work still receives worker handoff Category: ChatGPT / Bugs Tags: chatgpt, bug, 5.6, conversation, regression, projects @OpenAI_Support — I am requesting engineering review of a reproducible ChatGPT Plus execution-routing regression. This is not a subjective “5.6 feels worse” report and it is not based only on s…

---

_[View the full topic](https://community.openai.com/t/codex-rate-limits-discussion-thread/1378553)._