Is the 5-hour limit of Codex temporarily cancelled: shared quota, Banked Reset and usage view

Wed, 15 Jul 2026 21:57:00 +0800

If you don’t see the original “Restore in 5 hours” prompt in Codex or ChatGPT Work recently, don’t assume there’s something wrong with your account. OpenAI is temporarily adjusting the short-term usage window for some paid plans; however, this does not mean that all quotas are cancelled, nor does it mean that all accounts will automatically receive a saveable reset card.

As of July 2026, the safest way to judge is still to look at your Codex Settings > Usage panel and the topping out prompt. OpenAI’s help documentation confirms: Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents will consume the same agentic usage/credit pool when available; the available amounts are different for different packages, workspace roles and models. Codex Rate Description

Let me start with the conclusion: the temporary disappearance of the 5-hour window does not mean that there is no limit

This change is for the short-term 5-hour usage gate. Even if this window is temporarily no longer displayed on your interface, there may still be weekly quotas, credit limits, model-specific caps, workspace rules, and fair use protections; after reaching any limit, the product may still require you to wait for recovery, use other models, purchase additional credits, or upgrade packages.

So don’t interpret “no 5-hour countdown” to mean unlimited. OpenAI’s description of ChatGPT usage also clearly states that the limit will change with the package, workspace settings and product status. The recovery time displayed on the interface is the most direct basis for the current account. GPT-5.6 in ChatGPT

Why Codex and ChatGPT Work change together

Codex and ChatGPT Work are not two completely independent sets of counters. For plans with these features enabled, usage will be deducted from the same agentic usage/credit pool.

This has two practical consequences:

Running long tasks, research, or cross-application operations in Work may affect your available credit in Codex later.
Executing large warehouses, multiple rounds of agents, or high-inference tasks in Codex may also cause the remaining balance of Work to decrease.
When you encounter a limit, you should first look at the specific items in the Usage panel, not just whether a certain client still displays “5 hours”.

If you have just started using Work, you can first read ChatGPT Work’s long tasks and how to use the desktop, and then break down high-cost tasks into acceptable small stages.

How to understand GPT-5.6 Sol’s “reduce quota”

GPT-5.6 Sol is gradually available in Work and Codex for eligible Plus, Pro, Business, and Enterprise plans; these plans can also select Sol, Terra, Luna, and different effort tiers by product and permissions. OpenAI’s GPT-5.6 Release Notes

Some update notes describe Sol’s optimization as: when completing similar tasks, it is expected to consume less credit, which can approximately extend the available credit. But don’t think of “about 10%” as a fixed commitment. The actual deduction of Codex depends on the input, cached input, output token, task size, selected model and speed mode; the consumption may still increase significantly when the output is long, the context is large, or fast mode is used. Codex Rate Description

A more practical approach is:

Daily modification, positioning and small-scale review, prioritizing narrowing the scope of tasks and providing clear acceptance conditions.
Large refactorings require a plan, a documentation list, and risk points, and then have Codex execute them in stages.
When a long context is needed, try to reuse the same task context to avoid pasting the entire set of warehouse information repeatedly.
If you find that the credit limit drops abnormally fast, record the model, mode and task size used, and then check it against the Usage panel.

What is Banked Reset and who can get it?

Banked Reset (savable limit reset) can be understood as a “recovery coupon” that can be redeemed manually when needed: when you reach the resettable limit, you can use it to restore the corresponding Codex rate limit once, instead of having to wait for a fixed window to automatically restore.

However, it is not a default and permanent all-member rights. OpenAI’s current help document states: In some Plus and Pro recommendation activities, a successful recommendation may result in a banked Codex rate-limit reset; rewards, invitation upper limits, validity periods, and redemption conditions will vary depending on the activity and package. Eligible Business workspace members may also see an entry to invite colleagues. Using Codex with your ChatGPT plan

Therefore, please refer to the actual display in the account: if you do not see the reset number, it does not mean that the client is faulty, and you should not take online screenshots as proof that your account has been gifted.

Where to check the limit and reset the card

It is most reliable to check in the following order:

Open settings in Codex and enter the Usage panel to view remaining credit, current limits and available operations.
If there is a Banked Reset, open the usage summary in the profile menu; the official example will show the available amount like “1 reset available”.
If restrictions are triggered while the task is in progress, the current turn may continue to be completed within the scope allowed by the fair use rules; after the completion, you can choose the additional purchase amount, use the available reset, upgrade or wait for recovery according to the prompts.
If Business users cannot see the purchase or management entrance, first confirm their workspace role; some operations are only open to owners or administrators.

This is more efficient than repeatedly refreshing the desktop. Especially during temporary adjustments, product alerts, package qualifications, and workspace policies may be out of sync at any time.

A set of safety procedures after reaching the top

When you find that the 5-hour countdown has disappeared, or you cannot continue after the quota is used up, you can follow the following process:

1. 打开 Codex Settings > Usage，确认是信用额度、周额度还是模型限制。
2. 确认近期是否在 ChatGPT Work、Codex 或其他 Agent 功能执行过高成本任务。
3. 有 reset 时，核对它对应的提示和有效期，再在确有紧急任务时使用。
4. 没有 reset 时，选择等待恢复、降低任务规模、换可用模型，或按套餐权限加购额度。
5. 记录异常消耗的任务、模型和时间；若面板数据明显不合理，再向 OpenAI Support 提交截图和任务信息。

Regarding common short-term windows, weekly quotas and notification judgments, you can read them together with When does Codex Usage Limit reset: How to read 5 hours, weekly quotas and Reset notifications.

FAQ

The 5-hour limit is gone. Can I keep running tasks?

no. The temporary adjustment of the short-term window does not change other quotas, package rules, workspace policies and fair use protection. Please refer to the Usage panel and the top-out prompt.

Will the Banked Reset be automatically issued to every Plus, Business or Pro user?

It cannot be understood this way. Official public help documentation currently associates it with specific referral campaigns and account qualifications; rewards and availability vary by campaign and package. Only when the available reset amount appears in the account, it means that it can be redeemed currently.

Why is Codex affected if Work is used too much?

In plans where the feature is available, Codex and ChatGPT Work share the agentic usage/credit pool. They should be considered as different entries for the same job-type credit.

Can “GPT-5.6 Sol use 10% less quota” be used to plan the budget accordingly?

This ratio cannot be regarded as a guarantee. The task’s token structure, output length, model, and speed mode will all change the actual deduction; you should use your own Usage record and rate description for estimation.

summary

The core value of temporarily canceling the 5-hour gate is to reduce high-frequency work being interrupted by short windows; it does not turn the Codex into unlimited use. Manage Work and Codex as a shared quota pool, first look at the Usage panel, and then treat Banked Reset as a tool that “will appear only when conditions are met and can be reserved for emergencies” to avoid interrupting critical tasks due to misreading temporary policies.

GPT-5.6 Sol on KnightLi Blog