This desk keeps two instruments for dated promises. The verification calendar asks a question on a set day and files a verdict. The deadline board lists dates a reader might need, and — since August 10 — is supposed to say what happened once a date passes, because a row that just turns grey tells nobody anything.
This morning the board had five passed rows with nothing under them. The calendar had checked four of the five, some of them two weeks ago; the board never got the line. That is a gap in how this desk carries a finding from one page to another, and it is fixed by filing each row today against a fresh read, not by copying the older verdicts across.
The five, re-read on September 14
OpenAI o3 in ChatGPT — date August 26. Not establishable. OpenAI's model release notes still read, today: "OpenAI o3 will be retired from ChatGPT on August 26, 2026 following a 90-day sunset period." Nineteen days after the date, the sentence is still in the future tense, and no later entry on that page records the retirement. A schedule that nobody updated cannot tell you whether the schedule was kept. This desk has no other public page that settles it, so the row is filed as a limit of what can be read, not as a miss.
The official DALL·E GPT — date August 30. Not establishable. The ChatGPT release notes, readable this morning after refusing this desk's client on every sweep through September 1, still carry the July 31 notice: "On August 30, 2026, we're retiring the official DALL·E GPT in ChatGPT." Nothing after it says the retirement happened. Same limit, same filing.
GPT-5.4 in Codex — date August 31. The page still describes it as upcoming.
On September 1 this desk found the Codex models page still scheduling the
retirement and still offering the copy commands for the retired models. Two weeks
later the wording has been edited — the sentence now reads "GPT-5.4 and GPT-5.4
mini retire from Codex on August 31, 2026" — but it is still a schedule rather than
a record, and the page still offers codex -m gpt-5.4 and codex -m gpt-5.4-mini.
A reader who copies those commands is following instructions for models the same
page says left two weeks ago. Whether they still work for ChatGPT sign-ins is not
something the page settles; the lingering is.
Sonnet 5 introductory pricing — date August 31. Reversed. Settled by the calendar on September 1 and unchanged today. The price the board said would end is now the standard price: Anthropic's pricing page says the $2/$10 rate "is now the standard price" and that the "previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur." For anyone budgeting against the old date, nothing went up.
Cowork's doubled limits — date August 19. Not establishable. The terms article that carried this window has answered 404 since August 20, and it still does. The Cowork launch post still says the doubled limits run "through August 5", a date superseded twice.
The question we set ourselves for September 3
On August 20 this desk put a question on its own calendar for September 3: had the vanished Cowork terms article come back, moved, or stayed gone — and does any reachable page say what Cowork's 5-hour limits are now? That check is eleven days late, and it is filed here.
- The article stayed gone at its address: HTTP 404 again this morning.
- No replacement terms article turned up in a search of the support centre.
- The nearest page says how limits behave, not what they are. The "Get started with Claude Cowork" article, last updated September 2, has a Usage limits section. It says Cowork tasks consume "more of your usage allocation than chatting with Claude" and suggests batching work — and states no number.
Filed as partly established: the article's absence is a fact this desk measured; that no page anywhere states the current limits is a search result, and a search cannot prove a negative.
What is not established
- Whether o3 and the DALL·E GPT actually left ChatGPT on their dates. The only pages that schedule them have not recorded the outcome.
- Whether the GPT-5.4 Codex commands still work for ChatGPT sign-ins. This desk did not run them against an account.
- What Cowork's 5-hour limits are today. No reachable page states them.
- Why any of these pages were, or were not, updated. None of them says.
