Yes. A Codex usage reset was announced for paid Codex and ChatGPT Work users today, July 28, 2026, at about 03:09 UTC. That is the public answer. The account-level answer still depends on what your own Codex Usage screen or /status view shows. A global announcement is evidence that a reset was sent; it is not a guarantee that every meter on every account has already changed.
You can follow the current event and later announcements on the live Codex reset tracker. The first-party July 28 signal is the reset announcement from Codex engineering lead Thibault Sottiaux. Before starting a long task, use the verification method below instead of treating either link as a substitute for your authenticated account state.
The 60-second check: did Codex reset for your account?
- Open the authenticated Usage view. In Codex, open Settings and then Usage. If your client supports it, open
/statusas a second view. - Read both meters. Record the remaining percentage and reset time for the short window and the weekly window. Do not stop after seeing one meter at 100%.
- Refresh once. If you opened the page immediately after a public announcement, wait briefly and refresh one time. Avoid repeatedly reloading and interpreting every temporary display change as a new reset.
- Try a small, low-risk action. If both the meter and the client indicate capacity is available, use a bounded task before committing hours of work.
- Save the displayed reset time. It is more useful for planning your week than a prediction about the next global goodwill reset.
OpenAI‘s official guidance for using Codex with a ChatGPT plan tells users who are near or at a limit to check the Codex Usage page or the limit banner for the options available on their plan. Those options can include waiting, applying an available reset, adding credits, or changing plans. The authenticated display is therefore the decisive source for your account even when a public reset is real.
Why “Codex reset” can mean four different things
Most reset confusion comes from using one phrase for several different events. Separate them before deciding what to do next.
1. Your short usage window renewed
The short-window meter governs near-term activity. When it refills, you may be able to work again even though the weekly meter has not changed. A renewed short window is useful only while the broader weekly allowance still has room. If the weekly bucket is exhausted, seeing the short meter at 100% does not automatically mean a new long task will run.
2. Your weekly allowance renewed
The weekly meter is the larger planning constraint for sustained Codex use. Its displayed reset time belongs to your account state, not to a universal midnight shared by every user. The safest rule is simple: copy the exact time shown in your own Usage view and plan around that time. Do not replace it with a date inferred from someone else‘s screenshot.
3. A manual or global reset was announced
This is what happened publicly on July 28. A manual reset is an event initiated outside your normal scheduled rollover. It can move the public conversation quickly because many users see new capacity at roughly the same time. However, previous user reports have shown that the visible scope can be confusing: one meter may change before another, the next displayed reset date may move, or an account may not appear to match the broad announcement immediately.
That does not make the announcement useless. It means the announcement answers “was a broad reset sent?” while your Usage screen answers “what capacity can I use now?” Those are related but different questions.
4. You apply a banked reset
A banked reset is an account-specific option that eligible users may receive and choose to apply. It is not the same as a global reset, and a global reset announcement does not necessarily add another banked reset to your account. Before spending one, inspect both usage meters. If a public reset has already restored the exhausted bucket, using a banked reset may provide little immediate benefit. If the weekly bucket remains depleted and your account offers an applicable reset, the decision may be different.
What today‘s reset confirms—and what it does not
| Question | What you can conclude | Best evidence |
|---|---|---|
| Was a paid-user reset announced today? | Yes, on July 28, 2026. | The first-party announcement and dated tracker record. |
| Did my account receive usable capacity? | Only your authenticated meters and a bounded task can confirm it. | Codex Usage or /status. |
| Did both the short and weekly meters reset? | Do not assume so; inspect both. | The two separate meter rows. |
| Did I receive a new banked reset? | A global reset does not prove that. | Your account‘s available-reset count. |
| Will another reset happen later today? | No reliable promise follows from today‘s event. | No forecast should override your displayed reset time. |
This distinction matters because “100% remaining” is meaningful only when you know which bucket it describes. It also prevents a common planning mistake: assuming that a public celebration or compensation reset creates a permanent increase in the size of your plan. A reset replenishes a usage state; it does not by itself document a plan upgrade, a new weekly quota amount, or a change to future pricing.
How to interpret mismatched or confusing meters
If today‘s announcement and your Usage view do not agree, diagnose the state before escalating the workload.
| What you see | Likely interpretation | What to do |
|---|---|---|
| Short window is full; weekly is depleted | The two buckets are not giving you the same usable capacity. | Wait for the weekly time, use an available account option, or keep the task off Codex. |
| Weekly increased; short window still looks old | The display may not have refreshed uniformly. | Refresh once, compare with /status, then try a very small action. |
| Both meters look renewed, but a new task is blocked | The client, workspace, credits, or backend state may not match the display yet. | Capture the meter values and time, check the limit banner, and use official support if the mismatch persists. |
| The next weekly reset date moved | A manual reset may have started a new visible period rather than preserving the old date. | Use the newly displayed date for planning; do not assume unused percentage rolled forward. |
| A banked reset disappeared or did not help | The reset may have been applied to a different limiting state, expired, or failed to propagate correctly. | Stop retrying, record the before/after state, and contact official support. |
A screenshot is helpful evidence when the state is inconsistent, but redact account identifiers before sharing it publicly. Include the local time zone, both meter values, both reset times, client surface, plan, and the action that produced the limit message. That gives support a reproducible state instead of a vague “reset did not work” report.
Should you start a long Codex task now?
Use a three-way decision rather than a simple yes or no.
Start
Start the intended task when the relevant meters show enough room, a small action succeeds, and the work can preserve progress if capacity runs out. Keep changes committed or otherwise recoverable, divide the work into verifiable milestones, and avoid bundling unrelated goals into one long run.
Shorten
Shorten the task when capacity exists but the weekly meter is already meaningfully depleted. Ask for one concrete outcome: diagnose a bug, implement one bounded feature, review one diff, or produce a plan with a clear handoff. Long context, large repositories, high reasoning effort, tool-heavy work, and extended sessions can consume more of the shared pool than a small script. The official Help Center notes that usage varies with task size and complexity, model, and where the task runs.
Wait
Wait when the weekly bucket is exhausted, the display is oscillating, or a reset announcement has not changed the account state you need. Waiting for the exact authenticated reset time is more predictable than betting on another goodwill reset. If the task is urgent, use only an option your Usage page actually offers and whose cost or expiration you understand.
Remember that the usage pool can be shared
Codex may not be the only product drawing from the relevant agentic pool. OpenAI currently says Codex, ChatGPT Work, ChatGPT for Excel, and Workspace Agents share usage and credits when those features are available on a plan. A user who did little work in the Codex CLI can therefore still see more consumption than expected if another agentic surface was active.
This is different from ordinary ChatGPT chat, file-upload, image, or voice limits. Do not use an unrelated ChatGPT banner to calculate a Codex reset. Match the warning to the product and meter named in the authenticated Usage view.
Why the tracker and your account are both useful
The tracker answers a public-events question: was a reset announced, when did it happen, and what was the stated audience? It also lets you monitor future announcements without pretending it can see private account data. Your Usage view answers the private operational question: how much capacity remains for you and when your current meter says it will renew.
Use them in that order. Check the public signal for context, then use the authenticated meter for action. This prevents two opposite errors: dismissing a real global reset because your page has not refreshed yet, and starting a multi-hour task merely because a public post said “reset” while your weekly bucket remains exhausted.
Method and limitations
This article was verified on July 28, 2026 against the dated reset announcement, the independent tracker, and OpenAI‘s current Help Center guidance. It cannot access or inspect your account, and it does not claim that the July 28 event propagated identically to every plan, workspace, region, or meter. OpenAI can change limits, credit options, product pooling, and interface labels, so the in-product Usage display should override this dated article whenever they disagree.
Bottom line: Codex usage was publicly reset for paid Codex and ChatGPT Work users today. Check both your short and weekly meters before acting. If the required bucket is replenished and a bounded task succeeds, you can proceed. If only the announcement changed, wait for your authenticated state or use an account option shown by OpenAI—do not spend a banked reset or launch a long job on the public signal alone.
