Codex reset posts are easy to find and hard to use. A signal gets logged,
then re-shared two days later by someone who read it as live, and by then
"the weekly window reset" and "a one-off reset card" are sitting in the
same feed using the same word.
I wanted a single place to answer one narrow question: is this reset
claim new, or am I reading a two-day-old repost?
That turned into a small read-only tracker for public Codex usage-limit
reset signals — Codex Reset Radar.
Each event carries a date, a source, a confidence label, and a summary;
older events stay visible so the history doesn't reset itself every time
something lands.
The problem: "reset" is three different clocks
Reading these threads, the same word covers at least three events:
- the 5-hour usage window coming back,
- the weekly plan window replenishing,
- a one-off reset card that only some plans see.
Same word, three different clocks. A repost from three days ago reads
exactly like breaking news, and a signal that was true for one plan gets
read as "my quota should be back too."
What the log actually records
Every entry is a publicly reported signal, not a verified reset. As of
writing (October 2, 2026, UTC):
- 39 public signals logged since April 28, 2026
- across 33 distinct reset days (events ≠ days; several signals landed the same day)
- 4.8-day average gap between consecutive signal days
- longest wait so far: 37 days (April 28 → June 4)
- confidence split: 32 high / 7 medium
- latest public signal: 2026-09-30 01:48 UTC (GPT reset card for paid users)
"High confidence" means a strong public source said something — not
that your quota went back. For the "strong signal, nothing changed on my
account" case, the site keeps a reset-failed checklist: rollout lag,
plan scope, 5-hour vs weekly windows, and banked-reset confusion.
What it is not
- Not affiliated with or endorsed by OpenAI — unofficial by design.
- No OpenAI login, no API keys, no cookies, no session tokens. It never touches your account and can't see your plan or workspace.
- It does not predict your quota. Nothing public can. Your own Codex usage UI is the only account-level source of truth — the site says so in almost every state it's in.
If you'd rather pull the data than read it
Three read-only endpoints, no key required — a status file with derived
stats, a full event list, and an RSS feed:
the API page.
Fine for dashboards, status bots, and home-lab monitors. They carry the
same unofficial disclaimer as the site.
The full history lives at
the reset history page.
If something in the log is wrong, or the source ranking looks off, tell
me in the comments — that correction is worth more to me than the post
itself. Happy to walk through how the confidence labels are assigned,
too; that part took the most time to get right.
(Unofficial, not affiliated with OpenAI. Your Codex usage UI is still
the source of truth.)
Top comments (0)