Codex Usage Limit Reached? What to Do Next

When Codex says a usage limit has been reached, do not immediately upgrade, restart the task, or switch credentials. First identify the account, workspace, billing route, named allowance, Credit balance, and reset time that actually stopped the work. In an active Codex CLI session, run /status.

Then use only an option the account displays: wait for the scheduled reset, apply an available banked reset, buy an offered instant reset, add eligible ChatGPT usage Credits, change plans for recurring demand, or intentionally use a separately billed API-key route on a supported local surface. None of these options bypasses a real limit; each changes a different part of the billing or usage path.

Last verified: September 10, 2026. Codex plans, shared allowances, Credits, models, reset purchases, eligibility, and rates change. Follow the exact notice, current Usage page, checkout, and official documentation for the active account.

Editorial review: Agent.Space Editorial checked the OpenAI limit, reset, and Credit documentation used in this decision tree. The exact message and account controls remain the primary evidence for a real incident.

The 90-second Codex limit check

Before paying or retrying, check these seven items in order:

1. Account: Are you signed in with the email and sign-in method that owns the plan or Credits?
2. Workspace: Is Codex using your personal workspace, the intended Business or Enterprise workspace, or another organization?
3. Billing route: Does /status show ChatGPT sign-in or an API-key route? They are separate bills.
4. Balance type: Are you looking at included plan usage, ChatGPT usage Credits, a banked reset, or API balance?
5. Named limit: Does the message identify a model allowance, shared agentic pool, rolling window, weekly window, rate limit, or spend limit?
6. Shared activity: Did Codex, ChatGPT Work, ChatGPT for Excel, Workspace Agents, or another supported feature use the same shared allowance or Credit pool?
7. Service state: Is the message actually an authentication failure, API 429, spend cap, or service incident rather than a ChatGPT plan limit?

If the meter still looks wrong, save the exact message, timestamp and timezone, client, model, active workspace, and a redacted screenshot before contacting OpenAI Support. Never include a credential. OpenAI says Support does not manually reset ChatGPT or Codex usage limits; it can investigate incorrect counting or access that does not return after the displayed reset time.

Identify what actually ran out

OpenAI's Work and Codex usage guide says Work and Codex share the usage allowance included with the plan. For Plus and Pro, supported agentic features can also draw from the same agentic allowance and Credit pool. A lower-than-expected Codex meter can therefore reflect valid activity outside the current Codex screen.

Use this decision order while the task is blocked

1. Protect what already happened. Save changed files, note completed external actions, and identify the exact missing step. A retry can otherwise repeat work or spend more allowance.
2. Name the meter. In Settings → Usage, determine whether the blocker is the 5-hour allowance, weekly allowance, usage Credits, or another named limit. If /status shows an API-key route, leave the ChatGPT reset flow and inspect the owning API project instead.
3. If only the 5-hour window is exhausted, compare waiting with the options actually shown. The paid weekly-limit banner does not appear for a 5-hour-only limit. An eligible personal Plus or Pro account may still show Buy an instant reset in Usage settings after using some allowance, but that purchase refreshes both the 5-hour and weekly windows and moves the weekly schedule.
4. If the weekly window is exhausted, check saved benefits before buying. A promotional banked reset can be used later until it expires; a global reset would already have been applied automatically. If neither solves the current block, compare the natural reset, paid instant reset, and usage Credits using the remaining work and the live price.
5. If you buy an instant reset, assume it acts now. It does not wait for remaining usage, cannot be banked, and brings the normal weekly allowance forward instead of stacking extra entitlement on top. Your next weekly reset is seven days after the first Work or Codex request made after the reset—not necessarily seven days after payment.
6. Verify before restarting. Reopen Usage, confirm the account, workspace, 5-hour meter, weekly meter, Credit balance, and reset date. Resume only the bounded step that is still missing.

This order separates recovery from purchasing. It also prevents a ChatGPT reset from being used as an attempted fix for an API 429, authentication error, service incident, or missing model access.

Choose one official recovery path

Wait for the displayed reset

Choose this when the task can pause and you do not want another charge. Use the time shown for the active account and allowance; another user's schedule may not apply.

Apply a banked reset if one is available

OpenAI's banked Codex reset documentation defines it as a one-time promotional reset saved to an eligible account until it is used or expires. A full reset refreshes eligible 5-hour and weekly Codex usage windows and changes the weekly reset date.

It is not purchased credit and does not permanently increase the plan limit. An automatic or global reset is applied directly and cannot be saved. The banked Codex reset guide explains the exact Settings → Usage flow, eligibility boundaries, and missing-reset checks.

Buy a paid instant reset only if the schedule change fits

OpenAI's paid weekly Work and Codex reset guide says this purchase is currently limited to eligible personal Plus and Pro accounts. It is unavailable on Free, Go, Business, Enterprise, and Edu, and availability varies by account and billing country.

Checkout shows the price and payment method. When it succeeds, both the 5-hour and weekly Work and Codex allowances refresh immediately, even if usage remains. The reset cannot be stored or scheduled, and OpenAI says these purchases are generally non-refundable. Buy it only after comparing the natural reset time, remaining work, Credits, and the new weekly schedule.

Add eligible ChatGPT Credits

OpenAI's Flexible Usage Credits guide says included usage is consumed first, then supported activity can draw from an eligible usage-credit balance. If the account offers purchases, use Settings → Usage in ChatGPT or Usage & Billing in the Codex app.

Check the live rate card, payment method, automatic reload, expiry, and spending limit. ChatGPT usage Credits are a separate pay-as-you-go balance: they do not refill the included plan allowance, create a banked reset, or fund the API organization.

Change plans for recurring demand

Upgrade only when representative work repeatedly exhausts the correct meter. One unexplained spike is a reason to inspect the workload, not proof that a higher plan is required.

Use a separate API route intentionally

On a supported local Codex surface, an API key moves usage to the API organization or project that owns the credential. This creates a different bill, model-access policy, budget, and data-control path. It does not extend the ChatGPT allowance or reproduce every cloud-only feature.

The Codex pricing guide compares all of these routes without treating their balances as interchangeable.

Protect the current work before retrying

A limit can appear after a task has already changed files or completed an external action. Before another turn:

• inspect changed files and the current diff;
• review validation output and pending approvals;
• note external actions that may already have happened;
• save the exact point where the task stopped;
• retry only the missing, bounded step.

Restarting the app, clearing the conversation, or repeating the full prompt does not restore usage. It can spend more capacity or repeat a side effect.

Reduce avoidable usage after capacity returns

If the allowance feels smaller after switching to Astra, first use the quota-consumption breakdown to separate allowance changes from rates, cache composition, and repeated attempts. It includes a worked example and a way to measure accepted-task cost.

These practices do not restore an exhausted allowance, but they can reduce waste in the next window:

• give one bounded outcome with files, constraints, and acceptance checks;
• remove irrelevant logs and duplicate context;
• keep repository instructions focused;
• disable tools and MCP servers the task does not need;
• choose a suitable available model for routine work;
• inspect partial output before retrying;
• compare accepted tasks rather than prompt counts.

Usage depends on the model, execution location, task complexity, context, reasoning, speed, tools, retries, and shared activity. A one-file edit and a long repository migration are not equivalent messages.

What Agent.Space changes—and what it does not

Agent.Space can preserve project files and Sessions and provide a clear handoff between supported Agents. The guide to how Agent.Space works explains that Workspace and Session continuity separately from an official model allowance.

Agent.Space does not refill an OpenAI allowance, convert ChatGPT Credits into API balance, authorize a borrowed account, or bypass OpenAI restrictions. It is a separate product and billing decision.

If a separate managed path fits your workflow, review current Agent.Space plans, then create an Agent.Space account. Register and pay only when the displayed plan, model access, and Workspace workflow fit the work you need to do.