You’re Out of Codex and Work Usage: How Codex Resets Work

The short answer

“You’re out of Codex and Work usage” means the shared allowance on your plan is exhausted, or that the workspace credit gate says it is. It is not an account restriction, and switching models does not refill a shared pool. OpenAI’s help center says Work and Codex draw from one allowance, that the allowance is not a fixed count of messages, and that your next step depends on which meter hit zero and what the limit notice offers your account.

Four paths usually exist: wait for the refresh time shown in Usage, redeem a banked reset, buy a paid instant reset (only on eligible Plus and Pro personal accounts), or continue on credits. They are not interchangeable. A banked reset is a free, saved, one-time item that OpenAI grants selectively. A paid instant reset is bought on the spot and pulls your weekly allowance forward. A global or automatic reset is applied to selected accounts without any action from you. Each of them changes your next refresh time in a different way, and that is where most of the confusion comes from.

This guide is built from six OpenAI help-center articles, the public GitHub issue tracker for the Codex repository, and posts from OpenAI’s official accounts. Everything is labeled with its source, estimates are called OpenAI’s estimates, and nothing here predicts when the next reset will land, because OpenAI’s own documentation says free resets have no guaranteed availability or schedule. For the model pricing side of the same story, see the sibling guide GPT-6 Astra vs Sol vs Luna: pricing and usage; the per-token price table is deliberately not repeated here.

What the message actually means

The wording you see varies by surface and by cause. A report in the Codex repository quotes two variants side by side: Codex shows “You’re out of Codex and Work usage” with the line “Your workspace is out of credits. Add credits to continue using Codex and Work,” while Work itself shows “You’re out of Work usage for now.” Older reports use “You’re out of Codex messages” and “You’ve hit your usage limit.” These are different strings from different eras of the product, so the useful move is to read the second line of the notice and then open the usage page rather than treating the headline as a diagnosis.

OpenAI’s plan-usage article states the shared nature of the pool plainly: Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents use a shared allowance and credit pool where those features exist on your plan. Regular Chat conversations are not part of it. Voice has its own wrinkle. Task work started through Voice inside Work or Codex on the desktop app draws from the same allowance, while connected Voice time is metered separately where flexible pricing applies, and ordinary Chat Voice has separate limits and does not consume Codex usage.

One behavior matters for long runs. The help center says that if you reach a usage limit during an active turn, Codex can continue working on that turn, subject to fair-use limits, and that you should check the usage page or limit banner after the turn ends. In other words, hitting zero mid-task does not necessarily kill the task in flight, but it does block the next one. The same article tells you where to look: Settings or the usage dashboard on desktop and web, and /status in an active Codex CLI session. It also says Support cannot reset your limits. Support can investigate counting errors or access that did not return after a stated reset time, and that is the full extent of it.

The allowances behind the message

Depending on the plan, usage limits apply over a five-hour window, a weekly window, or both. The Astra usage article spells out how the two interact. The five-hour limit sets how much usage is included in each window, and a new window starts when you send your first message in Work or Codex after the previous window ends. The weekly limit controls how much included work you can use across the weekly period. Where both exist, you need allowance left in both to continue, and you can reach the five-hour limit before five hours have passed, even with weekly usage remaining.

The same article makes a second structural point that explains a common surprise. On plans with both limits, the same usage counts against both, but the periods can end at different times. The banked-reset article gives its own example: a weekly period ending at noon while the current five-hour period continues until 3 p.m., so you could hold 100% weekly and 10% five-hour at the same moment. Neither meter is a countdown from the same starting point. Treat them as two independent clocks fed by one stream of usage.

Two caveats apply to that table. First, “eligible” is doing real work in the paid column: OpenAI says availability varies by account, including plan and billing country, and that holding the same plan as someone who sees the option does not guarantee you will. Second, the Pro rows reflect the help center as of the verification date, and the Pro plan structure has been changing. OpenAI’s About ChatGPT Pro tiers page says that subscribers who held Pro 200 at any point from September 22 through 10 a.m. PT on September 29, 2026 can use the previous included allowance through October 29, 2026 while the subscription stays active, after which it moves to a lower included allowance at the same $200 price. The page gives no absolute allowance figures, so none are quoted here.

What burns usage faster

OpenAI is explicit that the allowance does not correspond to a fixed number of messages or tasks. What a window covers depends on the model, the task, and the settings. The Work and Codex article names the levers: larger inputs and outputs, higher reasoning settings, Fast mode, and tasks with multiple steps all increase usage, and different models can consume different amounts for the same task. The plan-usage article adds where the task runs and which tools it uses, and notes that a long-running task can use substantially more than a short request. Cloud tasks may use more allowance than local messages.

OpenAI also publishes ranges for Plus and Standard Business. These are OpenAI’s own estimates of local messages per five-hour window, not guarantees, and the article repeats that usage varies with task, model and settings.

The spread inside each row is wider than the spread between neighbouring rows, which is the honest reading of the table: the model is one input and the task is another. Reasoning level is the setting OpenAI addresses most directly. Its guidance is that lower effort is a useful starting point when you want your allowance to go further, that higher effort can use more of the allowance without always producing a better result, and that Astra at Low effort can outperform Sol at High. It also says increasing reasoning effort cannot supply missing files, connected apps or permissions, so a bad result caused by missing context will not be fixed by spending more.

Model choice has moved since the Astra launch. In OpenAI’s October 26 announcement of GPT-6.1 Terra, the company said that in Work and Codex the usage allowance goes further with GPT-6.1 Terra than with GPT-5.6 Terra, and far more than with GPT-6.1 Sol or GPT-6 Astra, while pointing in-depth research and complex coding to Sol or Astra. A separate post from Tibo (@thsottiaux) the same day put the ratio at 2X more usage than 6.1 Sol across API and subscriptions. That second figure is a social post, not help-center documentation, so treat it as directional. For model-by-model background see the GPT-6 Astra complete guide and the GPT-6.1 Sol guide; for plan tiers and Ultrafast multipliers see the Ultrafast and Pro 500 plans guide, and for the cheap end of the market see cheapest AI coding models in 2026.

Three kinds of reset

OpenAI’s “How resets work in Codex” article separates resets into three types, and the separation is the single most useful thing to internalize. They share the word reset, but they differ in who triggers them, whether they can be saved, whether they cost money, and whether they leave a trace in your history.

All three are one-time actions separate from your regular scheduled refresh. OpenAI says they do not permanently increase your plan’s limits or add usage credits, and that free banked and automatic resets are discretionary promotional benefits with no guaranteed availability or schedule. A banked reset is also not cash, API credit or a transferable balance. The practical consequence is that an unused usage meter at 100% is the normal state of an account that has no reset history, and the absence of a reset in your account is not a malfunction.

Using a banked reset

To redeem one, open Settings > Usage & billing in the desktop app, or Usage settings on the web, select the usage summary that shows a reset such as “1 reset available” or “Full reset,” review what it covers and when it expires, and confirm. The mobile app has neither the controls nor the history. In the Codex CLI, enter /usage and choose Redeem reset; if your CLI version does not show that option, use the desktop app or web. On desktop, Use reset redeems a saved reset in Available, while Buy an instant reset opens the separate paid offer, and a paid purchase never spends a banked reset.

There is a guardrail that explains a message many people hit. From Usage settings, at least one allowance the reset covers must have less than 80% remaining. If every covered allowance has 80% or more left, the page tells you “Your usage does not need a reset right now,” and the reset stays in your balance until you use it or it expires. The same logic applies to how a reset is counted: it is marked as used only when it restores at least one covered allowance. A reset that restores nothing is not consumed.

A full banked reset restores the main Codex allowances on your current plan, which means the monthly allowance on Go, the 5-hour and weekly allowances on Plus and Business Standard, and the weekly allowance on Pro 100, Pro 200 and Pro 500. Each covered allowance below 100% goes back to 100%, and an allowance already at 100% is not restarted. Using the earlier example, redeeming at 100% weekly and 10% five-hour restores the five-hour allowance and leaves the weekly period running. Separate model or feature allowances such as image generation, file uploads and regular Chat Voice are not touched.

The desktop app also has an option, Allow your assistant to use resets, that lets Codex redeem a banked reset on your behalf. It requires your main 5-hour or weekly allowance to be at 10% or less remaining. If a reset disappeared without you clicking anything, the help center tells you to check History and your recent conversation for a request or confirmation, and to turn that option off if you want manual control only. History lists two event types, Reset received (a banked reset was added, which does not mean your usage refreshed) and Reset used. A “received” entry may already have been used or may have expired, so Available is the list that tells you what you can still redeem.

The paid instant reset

OpenAI’s paid-reset article describes a product that is narrower than it first sounds. Eligible Plus and Pro personal accounts on ChatGPT web and the Codex desktop app can buy an instant reset from Usage settings before reaching a limit, or from an in-app offer after reaching the weekly limit. A completed purchase immediately restores both 5-hour and weekly usage. It is not available on Free, Go, Business, Enterprise or Edu plans, and availability varies by account. The article does not state a price; it says to review the price and payment method in checkout, so any number you see quoted elsewhere is not from this source.

The detail that surprises people is the schedule change. A paid reset pulls your normal weekly allowance forward rather than adding a separate entitlement. Your new weekly period starts with your first request in Work or Codex after the reset is applied, even if you had usage remaining, and your next automatic weekly reset is scheduled seven days after that first request, not necessarily seven days after payment. OpenAI says this shift is expected. Because the reset applies immediately, cannot be saved or scheduled, and does not wait for the rest of your allowance to run out, the article’s own advice is to wait to purchase if you want to use up what you have first.

Two availability rules prevent wasted searching. The desktop Usage settings option is hidden while both your 5-hour and weekly allowances are fully unused, and it appears after you have used some of either. The in-app banner is different: it is triggered by an exhausted weekly allowance, not by reaching only the 5-hour limit, and credits available to continue usage may affect whether it appears. Not seeing a banner after a five-hour limit does not mean you cannot buy a reset, so check desktop Usage settings. On money, OpenAI says it generally does not offer refunds for weekly limit reset purchases, that usage credits are non-refundable except where required by law, and that a receipt in Settings confirms a purchase without by itself confirming the reset applied. If usage was not restored, contact Support from the account that was charged.

Why your reset time can change

Banked, global and paid resets can change the next refresh time of the allowances they touch without changing your subscription billing date. When a reset restarts a usage period, the new schedule replaces the old one. OpenAI’s example: a weekly period due to end on Friday that restarts on Tuesday refreshes next on the following Tuesday, and the Friday refresh no longer applies. The help center’s instruction is to read your next refresh time from Usage rather than calculating it from when you clicked, paid or saw an announcement, and it warns that before a new period starts, the displayed time may be an estimate that can change.

A global reset can land on top of a banked one. If you redeem a banked reset at 10:00 a.m. and use 5% of the refreshed allowance, a global reset at 10:30 a.m. takes it from 95% to 100%. The two do not stack into 200%, the banked reset you spent is not returned, and unused banked resets stay in your balance with their original expiry, because a global reset does not consume them. A global reset can also change your allowance and next refresh time without creating any History entry, so a changed date by itself cannot tell you which reset applied.

If you plan work around a refresh time, the safe habit is to treat every displayed time as the current schedule rather than a promise. A change in that date alone does not mean another banked reset was used, and OpenAI recommends not redeeming or buying a second reset just to test whether the first one worked. Usage can take a short time to update; refresh or reopen the page and confirm the account, workspace and limit you are viewing.

What OpenAI said publicly

The help center describes mechanics. The record of when resets were actually announced sits in posts from OpenAI’s accounts, and the pattern is useful context for what “no guaranteed schedule” means in practice. Five posts are embedded below, each checked to exist on October 28, 2026. They are history, not a forecast.

On July 10, 2026, OpenAI’s Tibo (@thsottiaux) wrote that usage limits had been reset across Codex and ChatGPT Work, with another reset coming later that day, and listed what the team said it got wrong at launch, including making it too easy to use the highest-compute settings without making the impact on limits clear. It promised to make usage and reset timing much more visible.

On June 12, OpenAI’s developer account announced rate-limit banking: invite a friend to Codex and when they send their first message you both bank a reset to use when you need it, rolling out to Go, Plus, Pro and Business. That post is the origin of the referral-based banked resets the help center now describes.

For the GPT-6 Astra launch, OpenAI’s Tibo (@thsottiaux) said on September 3 that one banked reset would be given for every day a paid-plan user did not have access to Astra. The help center’s account of the offer is more specific: eligible existing Plus, Pro and Business subscribers in good standing received a banked reset on September 3 and again on September 4, with account-creation cutoffs of 10 p.m. PT and 8 p.m. PT respectively, and on September 7 an automatic reset refreshed eligible usage limits without adding a saved one. OpenAI adds that individual access issues after launch do not qualify for another reset.

On October 20, the same account addressed reports of usage draining faster than expected over a weekend. It said about 40,000 accounts had an inconsistency in which usage limits applied, so the usage counted did not match the selected plan, that the issue was corrected, that affected users were being contacted individually, and that credits would be added to their accounts. That is a statement of a specific, bounded counting error. It does not describe the contradictory-display reports covered further down.

The most recent post in this set is from October 28, when OpenAI’s Tibo (@thsottiaux) opened with a joke about iCloud plugins, then wrote “today is a reset,” in a run of daily posts numbered by day. In the same timeline sample pulled for this guide, posts on October 20, 22, 23, 25, 26 and 28 each mention a reset. Counting those posts is not the same as counting resets applied to your account, since the help center says automatic resets go to selected accounts and limits, so this guide does not assert how many reached any particular user.

Which allowance is empty? A decision tree

The fastest way to avoid spending money on the wrong fix is to identify the meter first. The tree below follows the help center’s own order: check the usage surface, find the empty allowance, then pick among the options shown for your account. Every branch ends in an option OpenAI documents. None of them involves switching models, because the Astra usage article states that switching models does not restore allowance in a shared usage pool.

START: Codex or Work shows "You're out of ... usage"
|
+-- 1. Open Settings > Usage (desktop) / Usage settings (web), or run /status in the CLI.
|      Right account? Right workspace? Refresh once.
|
+-- 2. Which meter is at 0%?
|     |
|     +-- 5-hour meter only (Plus, Business Standard)
|     |     -> Wait for the window to end, OR redeem a banked reset,
|     |        OR buy an instant reset (Plus/Pro personal), OR use credits.
|     |        Weekly meter is untouched by waiting.
|     |
|     +-- Weekly meter (any plan with a weekly allowance)
|     |     -> Wait for the shown refresh time, OR redeem a banked reset
|     |        (needs a covered allowance under 80% remaining),
|     |        OR buy an instant reset from the weekly-limit banner,
|     |        OR use credits.
|     |
|     +-- Monthly meter (Go)
|     |     -> Wait, or redeem a banked reset if one is listed.
|     |        Paid instant resets are not offered on Go.
|     |
|     +-- Nothing is at 0% (weekly shows 100%, or credits show a balance)
|           -> Go to section "100% shown but blocked".
|
+-- 3. Message says "workspace is out of credits"?
|     -> Business/Enterprise: the workspace owner refills credits.
|        Personal plan: check credits balance and the account in the picker.
|
+-- 4. Banner mentions images, files or Chat Voice?
      -> Separate limits. They do not consume Codex usage and Codex ignores them.

Read the 5-hour branch carefully if you are on Plus or Business Standard. Waiting is the only option that costs nothing and leaves your weekly clock alone. A paid instant reset on that branch also restarts the weekly period at your next request, which is a larger change than the problem you were solving. If the weekly meter is the empty one, the compare-and-choose step is between waiting until the displayed refresh, spending a banked reset if you hold one, paying for an instant reset, or continuing on credits, and the right answer depends mostly on how many days remain until your displayed refresh time.

100% shown but blocked

A recurring class of reports on the Codex repository describes the usage page and the execution gate disagreeing. The most detailed recent one, issue 38892, opened on August 16, 2026 and still open when checked, describes a Plus workspace whose Usage page showed 100% weekly remaining and 1,000 credits while Codex returned “You’re out of Codex and Work usage… Your workspace is out of credits” and Work returned “You’re out of Work usage for now.” The reporter saw the same result in the macOS app and in a browser, and restarting did not help, which is why they suspected a server-side entitlement or credit-ledger mismatch rather than a stale client. That diagnosis is the reporter’s hypothesis, not a confirmed cause.

It is not alone. Issue 38354 (August 13, open) describes a Plus account with the dashboard at 100% weekly remaining and CLI /status agreeing, yet Codex returned “You’ve hit your usage limit.” Issue 32680 (July 13, open) reports /status showing 83% of the weekly limit left while every prompt was blocked, with an error asking an existing Plus user to upgrade to Plus and a reset date that disagreed with the one in /status. Issue 35047 (July 23, open) describes the analytics surface showing contradictory pools while a global gate stayed blocked. Issue 27598 (June 11, open) is a Business account blocked with “workspace is out of credits” while the panel showed remaining quota, and issue 12299 (February 20, open) is the oldest in the set, reporting a block at 10% remaining.

What can be said safely: these are user reports, they span at least five months and several plan types, all six were still open when checked on October 28, and the one comment on issue 38892 in the data pulled for this guide is an automated duplicate-detection note rather than an OpenAI response. What cannot be said is how many accounts are affected, what the root cause is, or whether any individual case has been fixed. OpenAI’s separate October 20 statement about roughly 40,000 mis-limited accounts concerns a different and acknowledged problem, and nothing found ties it to these reports.

The help center offers a procedure that fits this situation. Confirm you are viewing the correct account, workspace and usage limit, then refresh or reopen the usage page. Do not buy another reset or redeem another banked reset to test a theory. The reporter on issue 38892 pointed out the money risk themselves: a visible credit balance can push people to buy credits they may not need. If the numbers still look wrong, OpenAI asks you to contact Support with the limit or credit balance you are questioning, the reset time shown, a screenshot with sensitive information removed, the Codex client and model, and the time of the event including your time zone. Business and Enterprise members who see “workspace is out of credits” should also ask the workspace owner to check credits, since that message is about the workspace gate, which is a different control from a personal weekly meter.

Claim and evidence table

Each row ties a claim from this guide to its source and says how firm the evidence is. “Help center” means an OpenAI help article read on October 28, 2026.

Practical checklist

• Read the meter before the headline. Open Usage settings (desktop or web) or run /status in the CLI and note which allowance is at zero and the refresh time shown.
• Check the account and workspace. Personal and workspace entitlements are separate, and OpenAI lists a wrong account or workspace as the first thing to rule out.
• Look at Available before buying anything. A banked reset you already hold is free. Remember the 80% rule: it will not redeem if every covered allowance is above that.
• Pick the cheapest option that fits. Waiting costs nothing; a paid reset changes your weekly schedule; credits are pay-as-you-go and drawn after included usage.
• Do not stack tests. Never redeem or buy a second reset just to see whether the first worked. Usage can lag; refresh first.
• Decide on assistant redemption. If you do not want Codex spending a banked reset, turn off Allow your assistant to use resets.
• Lower the burn rate. Pick the model and reasoning level before a big task, review results before increasing effort, and avoid Fast mode when you are rationing.
• Keep evidence if the numbers are wrong. Screenshot the usage page and the error, note the client, model and time zone, and send it to Support.
• Do not plan around rumors. A reset post is not a schedule. Use the refresh time in your account.

Limitations

This guide covers documented behavior as of October 28, 2026, and several things are outside it. OpenAI’s help articles are updated frequently, and the three reset-related ones carried “updated” stamps of hours or days when read, so details such as plan names, covered allowances and eligibility can shift. The developer pricing page at developers.openai.com/codex/pricing showed older model and plan content when checked, so none of its figures are used here. No absolute token or message allowance for any plan is published in the sources read, which is why none appears. The paid reset price is not stated in the help article. Reset counts, daily-reset series and plan-cut claims circulating on social media are third-party and were not verified, so they are excluded. The GitHub reports are individual cases whose causes OpenAI has not confirmed in the data pulled.

Verdict

If Codex says you are out of usage, the right response is a short diagnosis rather than a purchase. Identify the empty meter, check whether a banked reset is waiting in Available, and compare waiting against the plain cost of an instant reset or credits, remembering that an instant reset moves your whole weekly period. The three reset types are not variations of one product. Banked resets are saved and redeemable, automatic resets arrive unannounced to selected accounts and leave no history, and paid resets are immediate and shift the weekly clock. Treat any displayed refresh time as current rather than permanent.

The sharpest edge is the gap between the usage page and the execution gate. When the page says 100% and the product says blocked, the documented advice is to verify the account, refresh, avoid buying more, and write to Support with evidence, while the public issue tracker shows the problem has been reported repeatedly and remains unresolved in the cases reviewed. Last verified October 28, 2026.

FAQ

Sources

Sources checked October 28, 2026.

• How resets work in Codex (OpenAI Help Center): reset types, redemption steps, 80% and 10% rules, History, timing, eligibility
• Paid weekly Work and Codex rate limit resets (OpenAI Help Center): eligibility, purchase paths, schedule change, refunds
• Managing usage with GPT-6 Astra in Work and Codex (OpenAI Help Center): 5-hour and weekly windows, message-range estimates, launch resets
• Using Codex with your ChatGPT plan (OpenAI Help Center): shared pool, active-turn behavior, Support limits
• ChatGPT Work and Codex (OpenAI Help Center): product split, usage structure, Voice
• About ChatGPT Pro tiers (OpenAI Help Center): Pro 200 grandfathering window, credits behavior
• Tibo, July 10; OpenAI Developers, June 12; Tibo, September 3; Tibo, October 20; Tibo, October 28 (X posts, embedded above)
• openai/codex issue 38892 and related issues 38354, 32680, 35047, 27598, 12299 (user reports, not OpenAI statements)

Related reading: GPT-6 Astra complete guide, GPT-6.1 Sol guide, Ultrafast, Pro 500 and Pro 200 plans guide, cheapest AI coding models 2026, GPT-6 Astra vs Sol vs Luna pricing and usage.