The Codex weekly allowance vanished in hours
As of 2026-11-09, OpenAI's 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; after that turn, check the Codex usage page or the limit banner for the options on your account: add credits, apply an available reset, upgrade, or wait for the reset time. OpenAI Support does not reset usage limits, but if you believe your usage was counted incorrectly you can contact Support so it can investigate. In more detail, as of 2026-11-06 the ways out OpenAI lists when your Codex allowance is gone are: wait for the reset, use a saved reset, or continue with eligible credits, depending on your account; eligible personal Plus and Pro accounts can also buy an instant reset, depending on account and billing country; switching models does not restore allowance in a shared usage pool. Since 2026-10-29, Pro 100 and Pro 500 get twice their previous Work and Codex allowance, and the Pro 200 reduction was canceled, with Pro 200 subscribers on the lower allowance brought up to the same limit as the higher allowance. From December, Plus, Pro and Business allowances will be shared across ChatGPT and Codex and measured in tokens; Instant chat does not count, and as of 2026-11-09 OpenAI has still not published an exact date. Once the allowance is shared, the options OpenAI lists are to wait for the reset, upgrade your plan, or buy credits. In addition, OpenAI's Tibo Sottiaux mentioned a Codex reset in X posts on 2026-10-23 and 2026-10-25; the next reset date is not published, so the usage dashboard in your account is the reference. OpenAI's docs give local-message estimates per 5-hour period for Plus and Standard Business (not fixed message limits); Pro currently has no 5-hour limit, and weekly limits may also apply. OpenAI says usage depends on the model, context, reasoning, tool use, retrieval and caching; community reports describe a full week used up in hours. Work out where it went before deciding to downshift, top up, or switch.
Updated 2026-11-10
One key for Claude, GPT and leading Chinese models, billed per token, up and running in minutes.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Four observations
How fast a weekly allowance can go
OpenAI documents local-message estimates per 5-hour period for Plus and Standard Business, and says weekly limits may also apply; in community reports the same tier with a high reasoning tier can use a full week in hours.
Logged on ChatGPT Pro 200
A developer recorded burning the weekly allowance of ChatGPT Pro 200 in a single day on long agent orchestration.
What extra credits felt like
OpenAI's position is that you can purchase additional credits after hitting the limit and keep working without upgrading the plan; in community reports one person topped up $20 of OpenAI credits and it was gone in under 10 minutes, another added €100 and had single-digit euros left after an hour.
The most direct saving
OpenAI advises starting from the default reasoning effort and raising it only when a task needs deeper planning, and lists ways to stretch the allowance: control prompt size, limit source material, shrink AGENTS.md, use fewer MCP servers. The fix the community confirms most often is the same: drop work that does not need deep reasoning to a lower tier.
Where the allowance actually goes
OpenAI's own wording: usage depends on the model, context, reasoning, tool use, retrieval and caching, so prompt length alone is not a reliable estimate. The official reference figure is that a typical GPT-5.6 Sol task may use 2-15 credits (credits are the ChatGPT plan's scoring unit). Small scripts may consume only a fraction of the allowance, while larger projects, long-running tasks and sessions that need the agent to hold more context use significantly more per message.
Token consumption for the same model differs sharply across reasoning tiers. The recurring pattern in public reports: running the highest tier on work that never needed deep thinking — bulk renames, boilerplate, test loops. Done on a lower tier or a lightweight model, the quality difference is often imperceptible while consumption differs by an order of magnitude.
About "when does the weekly allowance reset"
As of 2026-11-09, OpenAI's Help Center says OpenAI may occasionally give eligible users a one-time reset of Codex or ChatGPT Work usage limits; eligibility and the limits it covers vary by offer, and future resets are not guaranteed. On 2026-10-23, OpenAI's Tibo Sottiaux said on X: “But yes… that’s a reset!” On 2026-10-25 he posted: “Day 20/ Rest and reset”. Neither post says what exactly was reset, for whom, or when the next reset will be.
The official docs give local-message estimates per 5-hour period and say weekly limits may also apply; the exact weekly numbers and reset times are the ones your usage dashboard shows, and a Codex CLI session can show what is left with /status. The next reset date is not published; the docs also say that if you reach a limit during an active turn, the agent can continue working on that turn. Community reports still disagree on the reset moment, so plan around the dashboard, not hearsay.
Timeline
Multiple developers publicly log high reasoning tiers exhausting a full week in hours.
First-person billing reports about extra credits cluster, from $20 to €100 of OpenAI credits.
The phenomenon persists: from 2026-11-06, Codex desktop app users filed issues in openai/codex (#59970, #59977) reporting usage that climbs after a resume or compaction, or forked threads that cannot reuse the parent thread's prompt cache. These are user-submitted reports, not an OpenAI conclusion.
Cut the burn rate first, then talk about switching
What is confirmed
The official document states that usage caps are given as estimates per 5-hour period and that weekly limits may also apply; after reaching the limit you can purchase additional credits to keep working without upgrading the plan; current limits and reset times sit in the usage dashboard, and a Codex CLI session can show what is left with /status; with an OpenAI API key you can run extra local chats, charged at standard API rates.
Public discussion contains many first-person logs as well: high reasoning tiers exhausting a weekly allowance in hours, and extra credits consumed within 10 minutes to an hour. These are user-reported billing experiences and are searchable.
Do not treat as fact
Claims that "the weekly allowance resets on Thursday for everyone" or that "the vendor quietly reduced allowances" have no official statement covering all tiers and no evidence at scale, so this page makes no such claim; as of 2026-11-09, OpenAI's Codex pricing page, in its GPT-6.1 Terra availability note, says overall Work and Codex plan limits are unchanged. The reset time shown in your own dashboard is the one that applies to you.
Two routes
Keep economising inside Codex
Downshift the reasoning tier, tighten scope, push simple work to lightweight models. No tool change, but the weekly ceiling is still the ceiling.
Switch model family when it runs out
One key, move to another family and keep working. Suits people already bouncing between two subscriptions. The cost is accepting different model habits.
Why topping up does not hold either
OpenAI's position is that you can buy additional credits after hitting the usage limit and keep working without upgrading the plan. But extra credits are drawn down at OpenAI's official price, so if your burn rate has not changed, any top-up burns at the same speed. One developer logged $20 gone in under 10 minutes; another had single-digit euros left an hour after adding €100. Bring the burn rate down first, then decide whether to top up — doing it the other way round means topping up forever.
What QCode can pick up
When the Codex allowance is gone, use the same QCode key to move to claude-sonnet-5, glm-5.3, kimi-k3, deepseek-v4-pro or qwen3.8-max and keep working, without buying a second subscription; it is billed per token, with per-model prices on /models. Some people also run a lighter model for execution and a stronger one to finish, pulling total consumption down.
Your Codex subscription keeps its own allowance and reset schedule, and QCode runs alongside it on a separate key. When you run short on QCode: Personal Plans users buy a booster pack, without a plan buy Pay-as-you-go Packs starting at $4 for a one-time quota you use within 7 days. Pack prices are on /pay-as-you-go.
FAQ
Why is the same work eating the allowance faster this week?
Most often the reasoning tier changed, or the model widened the scope on its own. Start by checking whether the highest tier is being spent on work that never needed deep thinking. OpenAI's Help Center also says that if Codex repeatedly shows "Context automatically compacted" without making progress, you should stop that task and start a new Codex chat with a short summary of your goal and the work already completed; starting a new chat does not reset your usage limits. If you believe usage was counted incorrectly, contact OpenAI Support with the limit or credit balance in question, the reset time shown, a screenshot, the Codex client and model, and when it happened, including your time zone (read on 2026-11-09).
Are extra credits worth it?
It depends on whether your burn rate came down. OpenAI's position is that you can buy additional credits after hitting the usage limit and keep working without upgrading the plan. Credits draw at OpenAI's official price, so an unchanged rate burns them at the same speed. Reduce the rate first, then top up.
When exactly does the weekly allowance reset?
As of 2026-11-09 the next reset date is not published, and OpenAI says future one-time resets are not guaranteed. The official docs give local-message estimates per 5-hour period and say weekly limits may also apply; the exact weekly numbers and reset times are the ones your usage dashboard shows, and a Codex CLI session can show what is left with /status. Using a full banked reset refreshes both your 5-hour and weekly Codex windows and changes your weekly reset date; after a purchased instant reset, the next automatic weekly reset is 7 days after your first Work or Codex request. Community reports still disagree, and there is no public official statement covering every tier, so use the time your own dashboard shows.
Does moving to QCode solve the weekly limit?
It solves the part that blocks you: "the allowance is gone but the work is not" — one key, switch model family, keep writing code, while the OpenAI-side weekly limit resets on its own schedule. The OpenAI docs also say all users may run extra local chats with an OpenAI API key, charged at OpenAI's standard API rates.
How is QCode billed differently?
By a daily limit rather than a weekly allowance, resetting at 00:00 Beijing time. If the day runs short you can self-serve a booster pack, which resets separately at 02:00 the next day. Predictable cost is the main characteristic.
Are lightweight models good enough?
It depends on the task. Some developers report no perceptible quality difference after moving execution work to a lightweight model; others hold that complex reasoning still needs a flagship model. The sensible pattern is layering: the lightweight model executes, the flagship model finishes.
Sources
The phenomena and figures cited here come from publicly searchable developer discussion between 2026-07-30 and 2026-08-20. On 2026-10-20 the OpenAI documentation (Codex pricing and usage limits, workspace usage limits, model notes) was checked as well; the official wording is the 5-hour period and the usage dashboard. Earlier official quota-reset announcements are historical events and do not describe the current state. Re-fetched 2026-10-26: learn.chatgpt.com Codex pricing and model docs, plus the X posts of OpenAI’s Tibo Sottiaux (@thsottiaux) on 2026-10-23 and 2026-10-25. Read again on 2026-11-06: the OpenAI Help Center ChatGPT Business release notes (2026-10-29 entry), Understanding ChatGPT and Codex usage limits, and Managing usage with GPT-6 Astra in Work and Codex. Read again on 2026-11-09: Using Codex with your ChatGPT plan, Understanding ChatGPT and Codex usage limits, Managing usage with GPT-6 Astra in Work and Codex, and the Codex pricing page; issues #59970 and #59977 in the openai/codex repository are user-submitted and were read the same day.
Allowance gone, work still due
One key across multiple model families, billed per token with per-model prices on /models. Out of quota? Personal Plans users buy a booster pack, without a plan buy Pay-as-you-go Packs starting at $4 for a one-time quota you use within 7 days.
Related reading
How to use Codex without a ChatGPT subscription
Two official sign-ins, API key switch steps, billing and feature limits, plus the QCode API key setup and Pay-as-you-go Packs.
The Codex 5-hour limit
How the short-cycle cap is calculated.
Pay-as-you-go Packs: No Subscription, 7 Days
One-time quota used within 7 days, no subscription; every pack lands on the same key and works in Claude Code and Codex.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
As of 2026-11-06, the third-party quota mechanics described here follow each vendor's official documentation and may change at any time. User experiences quoted are self-reported in public discussion and are not a promise about consumption rates on any tier; model availability is defined on /models.