What to do when GPT-6 Pro runs out: shared limits, fallback, and access checks

Scope of this revision

This article was first published on September 21, 2026. On October 4 it removed older universal exact message counts for personal Pro and an old community incident anecdote. This revision only corrects the explanation of limits. It is not a new model launch and does not establish that all older accounts switched to another set of limits simultaneously.

Before troubleshooting, record the plan, product entry point, model, and time of the notice together. Keep the original wording, then check the corresponding help page. Do not substitute historical screenshots, someone else's remaining usage, or an informal plan nickname for your own current conditions.

Distinguish the plan from the model first

Sol Pro and GPT-6 Pro are different options. Plus can use Sol Medium/High in Chat, but not Extra High/Pro. Free/Go use Luna.

For example, first note whether you mean Pro on the plan page or Pro in the chat model menu, then save the corresponding page. Saying only “I bought Pro but do not have Pro” leaves support unable to identify the missing option. Check the visible menu in a new conversation before submitting a long-running task.

Pro sharing and fallback

Pro 100's two Pro models share a weekly allowance. After Pro 200 reaches the GPT-6 Pro weekly cap, it falls back to GPT-5.6 Thinking at Medium. Switching back to Sol Pro still requires remaining capacity in its daily allowance and the combined daily allowance for both models.

Business Standard and Business Premium also share their respective Chat allowances between the two Pro models. Switching does not reset that allowance. Consult Business documentation for specific seat conditions rather than applying personal Pro rules.

When a fallback notice appears, first save the model names before and after fallback and the reset notice, then determine whether the current option can finish the remaining work. If a task requires a particular model, save its inputs and progress and continue after confirming availability. Repeated menu switching is not proof that allowance has recovered.

Chat, Work, and Codex: label the entry point for every usage record

Chat allowance cannot be converted into a number of remaining Work/Codex tasks. Their usage also varies with task size, input and output, reasoning, and speed settings. Codex usage with ChatGPT sign-in counts toward the plan; using your own API key incurs API billing.

For cost checks, keep three separate rows: ordinary conversations, agent tasks, and API projects. Record the actual entry point, observation time, and interface notice for each. The same email address does not imply the same bill. Verify the recording method with one small task before assessing batch processing. A successful task proves only that this run was possible, not that every model or tool is enabled.

A missing model: identify the specific version or permission difference

Update the client first, then confirm the product you are using. Enterprise/Edu administrators can open Admin console, select ChatGPT workspace, and go to Models > Test to search for a member and inspect effective access, roles, and settings. Model Test does not enable models, modify usage, or bypass seat eligibility.

The member should provide the current account, workspace, client version, and missing option. The administrator should record the test result and corresponding role. If the test shows access is allowed but the menu still lacks the model, retain observation times for both and report them to support. Setting a default model or seeing a colleague use it does not prove your own permissions have taken effect.

What to do after reaching the limit

Support does not reset ChatGPT or Codex allowances, but can investigate miscounted usage or access that remains unavailable after a reset.

Save the notice and time first, then check once after the displayed reset time. If no time is shown, do not invent a recovery deadline. Include the plan, entry point, model, version, observation time, and redacted screenshots in feedback so the issue can be reviewed. There is no need to submit complete payment information or experiment by canceling subscriptions or paying repeatedly.

Acceptance: prove the entry point fits the task instead of guessing universal counts

Choose a small task without sensitive data and record the selected model, final visible option, and whether the result meets the requirement. Save usage notices before and after. If you cannot distinguish request volume, task volume, or billing source, stop expanding batch tasks and complete the records before comparing.

Write a specific conclusion, such as “The personal account can select this model in Chat, but the weekly allowance notice has not reset,” rather than “the account is fully fixed.” Repeat the same checks when the menu or plan next changes. Old screenshots establish historical state, not that the same allowance remains available today.