01Overview

Experimental - X development data only

No unusual
complaint signal.

Public usage-limit complaints are within the expected range. This does not confirm that Codex itself is issue-free.

1200 posts checked
02Readout
Complaints 26 55.8 expected
Ratio 0.47× For context only
Last 2 hours 4 4.7 expected
Data Complete 1200 posts
03Trace

Trend

Complaint history

60 days
24h totals

183 122 0 Aug 18Sep 17Oct 16

Hover or tap a day for details. Use arrow keys when the chart is focused.

Complaints Expected Latest value may change
04Evidence

Recent complaints

What people are reporting

A small sample of recent posts classified as personal, abnormal Codex usage-limit complaints. Accounts and links are omitted.

Community evidence

Can someone explain why Codex is this slow? Out of every coding tool I’ve used, it’s the slowest — even on the highest tier. And that five-hour limit? What, work one hour then sit around for four??

Public X post

Codex. 200 pro plan. 2 hours gpt 6.1 sol high -33%. -100% weekly in 1 day easy. And 6.1 seems like more stupid. another scam prepared?

Public X post

@… I understand that not everyone can have your token budget, but simply managing coding agents with @… burns the new unified ultra quota in half a week, only slightly up from 2-3 days before. I mean, I already let the coding be done by Codex, Claude and co. Isn't this supposed…

Public X post

who needs reset from @… ? As I am at 3% and would love to see one for the weekend so I can play with codex. I have almost full claude but I would rather play with codex this weekend

Public X post

I’m trying codex for the first time and it’s completely unusable. One prompt and five minutes in and it’s hit the 5 hour usage limit. This is so lame.

Public X post

I got the Claude $200 plan and cancelled Codex $200 plan today. Main reason was the dumbness of GPT 6 astra in Web app building. You won't feel it unless you dig deeper. Another reason is the usage. Damn Check my profile I supported codex earlier but now I don't! #codex #claude

Public X post

These examples are community evidence, not proof of an outage.

05Method

Method

How the score works

The score sets the signal. The ratio is for context.

Version 2
01

Count complaints

Count unique people, not posts. Adjust the count when fewer than 1,200 posts are checked.

C̃24 = C24 × 1,200 / E24
02

Compare with history

Compare the count with the same time in past data. Recent days count more.

z24 = (C̃24 − μ24) / σ24
03

Add recent activity

The last two hours can raise the score. The extra boost is capped.

S = z24 + 0.25 × min(max(z2, 0), 3.0)
24h score-0.88 + Recency boost0.00 = Final score-0.88
Normal · no unusual signal < 1.5 Watch · possible issue ≥ 1.5 Probable · likely issue ≥ 2.5 and z24 ≥ 2.0 Strong · strong evidence ≥ 4.0 and z24 ≥ 3.0

The ratio is for context. It does not set the signal. Low sample counts limit or hide stronger signals. Repeated posts from one cluster cannot create a high signal. Engagement is not used.