Codex 28 · Day record · 20 Oct 2026

Day 16 of 28

Closed at 00:00 UTC on 21 Oct 2026 Counted only from activity timestamped inside this UTC day. Later votes, answers and experiences do not change it; only an entry in the corrections log can.

Revised 25 Oct 2026, 07:53 UTC after an entry in the corrections log. Corrections log →

Download card

The day’s change

Recorded update

Improvement

Day 16: Tibo says about 40,000 accounts had usage metered against the wrong plan's limits, now corrected, with credits for those affected

Tibo writes: "There were reports of usage draining faster than anticipated for some users over the weekend. We've identified approximately 40,000 accounts that had an inconsistency on which usage limits where the actual usage that was applied did not correspond to their actual selected plan." He says OpenAI has corrected the issue, is reaching out individually to these users, and will add credits to affected accounts: "Sorry we got this wrong." It was posted on 20 October at 17:10 UTC.

Read the whole note

This is a fix for a metering inconsistency on a set of accounts plus compensating credits for them, not a usage reset for everyone and not a change to the plan limits themselves. The post does not name the products or plans involved, the exact dates affected, how many credits each account receives or when they land, and it does not label itself a Day 16 delivery; it is recorded under the UTC day it was posted.

If your usage drained unusually fast over the weekend, check for a message from OpenAI and for added credits in your usage panel. Rantr has not independently verified the 40,000 figure, the fix or the credits.

Source ↗

Fewer than 10 votes by the close of the day; the tally is in the open data, not shown here.

Improvement

Day 16.2: Tibo announces 12-hour prompt caching in the API and a cache kept indefinitely for signed-in Codex and Work users

Tibo writes: "Day 16/ You can now cache for 12h in the API and for logged in users across Codex & Work we keep cache forever and you don't have any penalty for coming back to an old conversation or resuming after a few hours and are now guaranteed to hit the cache." He quotes OpenAI Developers: "You can now choose 12-hour prompt caching in the API", so shared instructions, tools and context stay cached longer and requests hours apart pay the lower cached-input rate.

Read the whole note

This is a caching change: an opt-in 12-hour cache for API requests, and, by Tibo's account, no cache expiry for signed-in Codex and Work users, so resuming an old conversation should not re-bill its context at the full input rate. It is not a usage reset and not a change to the stated plan limits, and the post gives no measured effect on how long an allowance lasts. It follows the Day 12 caching improvement. This is the post Tibo labels "Day 16/"; it was posted at 01:10 UTC on 21 October and is recorded under Day 16 as he names it, alongside the Day 16 metering fix.

To check it, resume a Codex or Work conversation after a few hours and compare the usage it consumes with a fresh one; API users can choose the 12-hour cache and look for cached input tokens in the usage response. Rantr has not independently verified the cache behaviour or its effect on usage.

Source ↗

Recorded after the day closed (21 Oct 2026, 01:13 UTC), so no votes fall inside the day.

Fewer than 10 votes by the close of the day; the tally is in the open data, not shown here.

Running totals

What changed since day 15

Totals since 5 Oct 2026, at the close of each day; today’s are so far. A total never goes down when a sample is recounted.

Replies on X

Replies to Tibo’s post published this day

A sample collected from X, not every reply. Each link opens the reply on X.

Replies dated this day in the collected sample: 134. Collected since 5 Oct 2026: 2214.

X replies are found by search and each one is checked against its public post on X. Last read: 26 Oct 2026, 23:46 UTC. Not every reply is read.

3 X replies from this day are not quoted: Rantr does not have their words or held them back after automated screening. They stay on X.

Nothing this day for: OpenAI release notes, published experiences, corrections.

Method

How this record is made

Each day runs from 00:00 to 24:00 UTC. The record counts votes, release check answers, experiences, “Same here” and replies timestamped inside the day. When the day closes the record is frozen; it is recomputed only after an entry in the public corrections log, or when the day’s update is recorded late. Rantr is a neutral record: it does not take a side.

Corrections log → Open data and method →