ChatGPT plan usage limits per app: diagnose caps before buying credits

An app can exhaust its own weekly ChatGPT cap while your account still has usage left. Check Settings → Usage before reconnecting or buying credits: login, permission to use your plan, the app cap and the overall allowance are different checks.

Originally checked 27 October 2026; source-checked 2 November 2026 against OpenAI’s app-usage help page. This page is a diagnostic worksheet for participating plan-sharing apps, not a promise that every tool with a GPT model has these controls. Find compatible tools in the ChatGPT app hub.

Did you authorize sign-in or actual plan usage?

OpenAI’s identity documentation separates account sign-in from permission to use subscription allowance. A connected account does not establish that you approved inference billing through the plan.

Before troubleshooting, write down the app name, active ChatGPT account, selected provider and exact error. Then identify the route:

What you see	What to check next	What not to assume
ChatGPT sign-in succeeded, AI unavailable	Plan-use consent and app eligibility	Login alone pays for inference
App reports a usage limit	App cap and overall reset information	The account must be empty
Credits exist but work is blocked	Credit permission and the app’s cap	A balance overrides every limit
Usage cannot be displayed	Service error and a later retry	Missing data means zero allowance
API-key provider selected	That provider’s billing settings	ChatGPT app caps govern API charges

OpenAI’s current help page lists Go, Plus and Pro for participating commercial plan-sharing apps, subject to the app’s own requirements. Its treatment of supported open-source tools is different. Do not use that commercial-app list to infer the entitlement of a particular Codex integration.

How does an app cap differ from your total allowance?

OpenAI defines the weekly app setting as a percentage ceiling against your overall weekly usage, not a reserved allocation. Manage it through Settings → Usage → App limits → Manage limits, then save the change.

Consider a hypothetical setup: you give a documentation app a 20% cap and a coding app a 40% cap. That does not create two wallets or guarantee that either app can spend its full share later. If another workload consumes the common allowance first, neither cap replenishes it. Conversely, a documentation app at its ceiling can be blocked while the coding app still works.

For a controlled check, pause unattended tasks. Record the app cap, the amount already used and the overall reset information. Run one small task, then compare the same fields again. Treat this as an observation exercise, not a benchmark: one run cannot establish a universal cost per prompt.

OpenAI’s pricing guide says model choice, context, reasoning, tool use, retrieval and caching affect consumption. It also warns against converting API token prices into estimates of included tasks. Keep a task description alongside the usage reading so a later comparison has context.

Why can an app have credits but still be unable to continue?

According to the app-usage help page, credit permission is separate, off by default and subject to availability. The app needs a 100% cap to use credits after included usage runs out; setting that cap alone does not authorize credits.

Treat the decision as a budget change. Our recommended troubleshooting order is to capture the error, confirm the intended limit and decide whether the task can wait. Only then consider changing credit permission. Review automatic credit purchases separately so a continuity fix does not become an unexpected spending policy.

For API-key routes, use the API provider’s billing controls instead. OpenAI’s authentication documentation distinguishes that route from ChatGPT sign-in. If you are using Hermes, its provider guide on BringYourSubs explains why the active provider and auxiliary routes need inspection too.

How do you disconnect an app without confusing it with signing out?

For connections offering the control, OpenAI directs you to Settings → Security and login → Login connections → Sign in with ChatGPT, then the app’s Manage connection → Disconnect. Confirm the action. Signing out of the app is separate from revoking its authorized plan access.

After disconnecting, do not immediately repeat the sign-in flow unless you intend to restore access. Record what you revoked and stop unattended jobs in the app as a separate precaution. Revoking a connection is not a substitute for reviewing information already sent to that provider.

What evidence should you collect before contacting support?

Prepare a short incident note: app and provider, approximate time with timezone, task requested, whether output arrived, the exact error and the relevant usage period. Include a request or job reference if available. Redact personal content and never attach passwords, tokens or credential files.

Send an app execution problem to the app maker; take a suspected ChatGPT allowance or recorded-usage error to OpenAI, with the maker’s findings if available. Repeated login attempts are not a usage-reset strategy. For a different problem, setting up an app in the first place, start with our Claude and ChatGPT subscription setup guide.