ChatGPT Usage Limits and Reset Times: The Complete Guide
By Daniel Mercer• 15 mins read
Published:
, Updated:
Understand ChatGPT usage limits, plan allowances and reset times, with practical advice on Work, Codex, voice, uploads, credits and interrupted tasks.
ChatGPT usage limits do not reset at one universal time. A text conversation, a Work task, a Codex session and a voice call can draw on different allowances, each with its own rules. Before waiting until midnight or buying more capacity, identify which limit you have actually reached.
This guide explains the rules documented by OpenAI as of November 6, 2026. It covers ordinary chat, agentic work, uploads, voice, research, credits and resets, with practical examples of how to read a usage warning. Account-specific counters remain essential: a plan name alone does not establish the exact allowance, purchase options or rollout available to you.
Start With the Message That Stopped You
“Limit reached” is a category of problem, rather than a complete diagnosis. Write down the feature, selected model, account or workspace, remaining allowance and any reset timestamp. Those details tell you whether the next useful action is waiting, changing an available setting, freeing storage or checking an administrator’s controls.
A model-limit notice concerns access to a particular model or thinking allowance.
A Work or Codex meter concerns agentic usage, which can be shared across those experiences.
An upload warning can concern the upload rate, saved-file storage or the individual file.
A voice warning concerns the relevant voice experience and its allowance.
A full conversation concerns context capacity, which is different from a time-based usage allowance.
For example, a rejected 600 MB document is not evidence that your text-chat budget has expired. Similarly, being unable to start another agent task does not establish that an available everyday chat model is exhausted. Treat the wording on screen as the starting point for diagnosis.
What Your ChatGPT Plan Actually Includes
Free and Go users have unlimited everyday text conversations, subject to abuse-prevention safeguards. That does not make every tool unlimited. OpenAI’s Free-tier FAQ distinguishes ordinary text from separately limited capabilities such as uploads, image generation and data analysis.
In regular Chat, Free and Go use GPT-5.6 Luna as that default rolls out. Plus includes Medium and High thinking; higher paid plans offer additional options, subject to access controls. OpenAI’s model guide explains that GPT-6 Pro has limits of its own, separate from Work and Codex. Plus access to Astra in Work does not include GPT-6 Pro in regular Chat.
Pro currently has $100, $200 and $500 monthly tiers. OpenAI’s tier documentation confirms that the previously planned reduction in $200 Pro Work/Codex usage was canceled. Pro 100 and Pro 500 received doubled Work/Codex allowances; lower-allowance Pro 200 accounts were raised to match the higher allowance, while existing higher-allowance accounts retained it. Plus was unchanged. Those changes do not mean every regular Chat allowance doubled.
For Business, Enterprise and Edu, identify your seat type and workspace settings before applying personal-plan advice. OpenAI’s pricing documentation describes flexible pricing and credits for eligible managed workspaces. A colleague’s access can differ because of their seat, role or billing arrangement, even when both accounts display the same organization name.
How Five-Hour and Weekly Resets Work
Work and Codex share an allowance. Where both five-hour and weekly limits apply, you need capacity remaining in both. OpenAI’s Astra usage guide states that a new five-hour window starts with your first Work or Codex message after the preceding window has ended. You can use up that allowance before five hours elapse.
For an illustrative example, a five-hour window beginning at 9:40 a.m. ends at 2:40 p.m., even if intensive work exhausts it by 11 a.m. A noon request does not restart that existing window. Your own meter supplies the applicable timestamp.
If the shorter allowance renews while your weekly allowance remains exhausted, you still cannot resume included usage. Read both meters instead of treating the first reset as your restart time.
Pro 100, Pro 200 and Pro 500 currently have no five-hour Work/Codex limit, but still have an included allowance. Banked-reset documentation identifies Go’s main Codex allowance as monthly, Plus and Business Standard as five-hour plus weekly, and those Pro tiers as weekly. These are agentic allowances, not a universal schedule for every ChatGPT feature.
Where to Check Your Remaining Usage
Start in Settings, then Usage. Depending on the client, the section may be labeled Usage & Billing. OpenAI’s Work/Codex guidance directs users to the current meter for remaining allowance and reset times. If you use the Codex CLI, its usage/status controls provide another way to inspect the relevant session and account information.
Read whether a percentage means used or remaining. A meter showing 30% used implies 70% remaining; a meter showing 30% remaining means most of that allowance has already gone. Also record the window label. “Weekly remaining” and “five-hour remaining” answer different questions.
When coordinating across time zones, write the date, time and zone together: “November 6, 14:40 UTC,” for example. Convert the actual timestamp rather than assuming a reset follows your local midnight. A subscription renewal date is a billing event; it does not establish the refresh time of every model and tool.
Keep a screenshot of an unexpected warning before changing settings. If the meter appears inconsistent, compare the same account and workspace across clients. A personal Pro subscription and a company workspace are different contexts for checking entitlement; an old screenshot from another workspace is poor evidence of an error.
Why a Message Count Is Only an Estimate
A two-line request and a repository-wide investigation are both messages, but they can require very different amounts of computation. Work/Codex allowance depends on the model, task and settings. This is why a fixed “messages left” prediction can disappoint someone whose next task is much larger than their previous one.
For Plus and Business Standard, OpenAI currently publishes these estimated local-message ranges per five-hour window:
GPT-6 Astra: 5–45 messages.
GPT-6.1 Sol: 15–160 messages.
GPT-6.1 Terra: 35–285 messages.
GPT-6 Sol: 15–150 messages.
GPT-6 Luna: 350–3,000 messages.
These pricing-page figures are estimates, not guaranteed caps, and they do not describe regular Chat limits. Cloud tasks can differ from local messages, and a weekly limit may bind first. Use the ranges to understand relative consumption, rather than to promise that a particular project will fit into a subscription.
For planning, measure a representative task: one chapter review, one coding issue or one research question with a defined output. Record the allowance before and after, the chosen model and whether tools were involved. A few comparable observations are more useful than dividing your plan price by an assumed universal message quota.
Fast and Ultrafast Can Spend Allowance Faster
Speed settings deserve a separate check when usage seems to disappear unexpectedly. OpenAI’s speed documentation specifies that Fast consumes included allowance at 2.5x the Standard rate, while purchased credits and Enterprise pay-as-you-go usage are billed at 2x.
Ultrafast supports Astra and GPT-6.1 Sol. It consumes included allowance at 8x the Standard rate, or purchased credits and eligible pay-as-you-go usage at 6x. Access depends on the plan and workspace: Pro $500 and eligible Enterprise/Edu arrangements can qualify. The usage multiplier describes consumption; it is not a promise that your complete task will finish that many times faster.
Choose acceleration for work where waiting matters. A deadline-sensitive correction may justify it; an overnight background task may not. Keep model, reasoning effort and speed as separate decisions. Choosing the highest value for all three without considering the task makes the remaining-budget estimate less useful.
Voice Has Its Own Clock
For Live in Chat, OpenAI’s Voice guide lists a rolling 24-hour period. Go and Plus have three hours; Pro $100 has 15 hours; Pro $200 has unlimited Live usage subject to abuse guardrails. Free access is limited and can change. Business Standard lists three hours and Business Premium 15 hours, with additional usage billed where supported.
A rolling 24-hour measure is different from everyone receiving a new allowance at midnight. If you used most of your time yesterday evening, do not assume it is all available again early this morning. Check the current notice before arranging a long voice-based session.
Voice in Work on web/mobile follows the plan’s voice rules while the task also consumes Work usage. Desktop voice is priced separately: the pricing page lists $0.05 per connected minute against the Codex budget, with the task’s model usage charged separately. Do not apply a Chat voice-hours allowance to desktop agent orchestration.
Uploads: Rate, Storage and File Size Are Different Limits
OpenAI’s upload documentation sets an upload-rate ceiling of up to 80 files per three hours, with Free limited to three uploads daily. Peak-time limits may be lower. Failed attempts can sometimes count, and ChatGPT does not currently display a complete remaining rolling-upload quota.
Saved-file storage is separately capped at 25 GB per user and 100 GB per organization. Check Settings > Storage. Waiting for the upload-rate window does not free storage. Deleting a conversation also does not necessarily delete a separately saved Library or Space file.
Documents and presentations can be up to 512 MB; text/document files also have a two-million-token cap. Spreadsheets are approximately 50 MB, depending on row size, and images are limited to 20 MB.
Inspect the failure before repeatedly retrying. If one oversized attachment fails, reduce or split that file. If storage is full, review saved files you no longer need. If the warning gives a retry time, use that time rather than generating more unsuccessful attempts.
Images, Data Analysis and Deep Research
Everyday text access does not guarantee more image generations or data-analysis runs. Use the specific feature’s warning and available counter. Avoid converting an old screenshot of someone else’s daily image allowance into a rule for every plan, region and account.
Deep research has a particularly important exception to calendar-month thinking. OpenAI’s research documentation says that plans with a fixed monthly allowance reset every 30 days from first use. The in-product counter shows remaining tasks. A monthly allowance therefore does not automatically mean the first day of the month or your card’s renewal date.
Before starting research, decide whether you need a multi-source investigation or a quick factual lookup. Specify the question, date range, required sources and final deliverable. That reduces the chance of spending a research task on a broad report you immediately have to commission again with a better brief.
A Full Chat Is Not a Used-Up Subscription
A conversation can run out of context space even when your account still has usage remaining. The Free-tier FAQ distinguishes this from a rate limit: a new chat can let a Free user continue without waiting. Context capacity includes the information the conversation must process, including incoming material and generated responses.
Starting again needs a handover. Save the goal, decisions, essential facts, unresolved questions and links to source files. Do not carry across every abandoned approach. A compact working brief makes it easier to continue accurately and helps you notice when the new conversation lacks something important.
Our guide to AI tokens explains the units behind these constraints. For longer agent workflows, context engineering covers how to maintain relevant working information. A fresh conversation can solve a context problem; it does not manufacture a new shared usage allowance.
Buying Credits: Pay for Supported Additional Usage
Usage credits are a pay-as-you-go balance for eligible features after included usage is exhausted. OpenAI’s personal-credit guide explains that Work, Codex and supported office tools can share an agentic allowance and credits. Plus/Pro users can check the purchase options in Usage settings. A limited group of Free/Go users also has credit purchases; availability is account-specific.
Buying credits does not upgrade the plan or increase its included allowance. Usage credits are also distinct from a ChatGPT wallet balance and API credits. If automatic reload is available, review its trigger, target balance and monthly spending limit before enabling it. Turning it on with a balance already below the trigger can cause an immediate purchase.
For a practical decision, compare the work still required with the next automatic reset. Buying capacity for one urgent, bounded deliverable is easier to assess than paying to keep an undefined investigation running. Check that the specific blocked feature can use the credits before you buy.
Banked Resets, Automatic Resets and Paid Resets
A banked reset is a free, one-time entitlement stored for later redemption; simply receiving one does not refresh usage. An automatic promotional reset applies directly to selected accounts and allowances. Neither should be treated as a scheduled benefit that every subscriber will receive.
For manual redemption of a banked reset, at least one covered allowance must have less than 80% remaining. That means more than 20% has been used, not that you must use 80%. Redemption is available through supported desktop, web or CLI controls, rather than mobile. Covered allowances below 100% are restored; already-full allowances are not restarted. Separate Chat image, upload and regular voice limits remain unchanged.
A purchased instant reset works differently. OpenAI’s paid-reset guide describes an account-dependent offer for eligible personal Plus/Pro subscribers. It applies immediately and cannot be banked or scheduled. It brings the normal weekly allowance forward rather than adding a separate entitlement.
The new weekly period begins with the first Work/Codex request after the paid reset is applied. The following automatic reset is seven days after that request, not necessarily seven days after payment. For example, buying on Friday and first resuming on Saturday anchors the new week to Saturday’s request.
Compare the checkout price, remaining allowance and next scheduled refresh before purchasing. If a completed purchase does not appear, refresh the correct account’s meter and contact Support with the receipt instead of buying another reset as a test.
Connected Apps Can Use the Same Allowance
Changing the interface does not always change the budget. OpenAI’s sign-in documentation explains that eligible partner apps can use your existing Work/Codex allowance when subscription sharing is enabled. Signing in and authorizing usage are distinct steps.
Check Settings > Usage > App limits for supported weekly percentage caps. A cap limits how much an app may consume; it does not reserve that portion for the app or create additional allowance. If another permitted workflow spends the shared pool, the capped app can still encounter the account’s overall limit.
Consider a simple budget example: setting a research app’s cap to 10% is a ceiling, rather than a guaranteed 10% waiting in reserve. Review partner usage before assuming that every unexpected drop came from the conversation currently open on your screen.
API Credits Are Separate, Even With the New Subscriber Benefit
The old blanket advice that a ChatGPT subscription never includes API credits is now incomplete. OpenAI’s subscriber-benefit documentation lists claimable API credits for eligible Pro and Business subscriptions. Pro tiers list $100, $200 or $500 per month; Business amounts depend on paid seats and workspace caps.
The benefit requires activation by linking an eligible subscription to an API organization you own. Eligibility includes a paid subscription at least seven days old and matching account email addresses. Issued credits follow the relevant billing-cycle expiry; unused credits do not roll over.
Crucially, those API credits do not increase ChatGPT or signed-in Codex limits, and the benefit does not support Codex usage through API keys. API requests also have their own rate limits and organization billing. Inspect API Platform billing for API-credit questions, rather than expecting a ChatGPT reset to resolve an API error.
What Is Changing in December
OpenAI has announced that, starting in December, Plus, Pro and Business usage will be shared across ChatGPT and Codex as Work’s agentic capabilities enter the main ChatGPT experience. The notice says usage will be measured in tokens, with Instant chat excluded from that allowance.
This is an announced transition, not a reason to describe every November allowance as already combined. The notice does not give a specific rollout day. Recheck Settings > Usage as the change reaches your account, particularly if your workflow moves between regular chat and agents.
What to Do When a Limit Interrupts Important Work
Save the exact warning and identify the feature, model, account and workspace involved.
Check every relevant meter, including a weekly allowance alongside any shorter window.
Preserve the unfinished deliverable, source links and a short handover before changing conversations.
Use an available lower-consumption option for suitable remaining work, without assuming it resets a shared pool.
Compare waiting, eligible credits, a reset or a plan change against the actual deadline and work remaining.
If access remains unavailable after the stated reset, check service status and contact Support with timestamps and evidence.
OpenAI’s model guidance says Support can investigate incorrect counting or access that has not returned, but cannot reset ChatGPT or Codex limits. Repeatedly reopening the app is therefore a diagnostic step at most, rather than a way to create new capacity.
For recurring work, build allowance checks into the schedule. Run a small representative task before a deadline-heavy day, keep task briefs specific, and separate essential output from optional exploration. Evaluate the usable result, rather than the number of messages exchanged; our agent evaluation guide explains how to check completion and reliability.
The most useful habit is to ask three questions whenever a warning appears: which allowance stopped this task, when does that allowance become available, and what permitted alternative can finish the remaining work? That approach remains useful even as plan names, models and reset options change.
