Less guesswork. Better AI spending.
Codex Usage Limits 2026: Allowance, Credits & Reset Times
Check the Codex dashboard and /status, identify the active billing path, plan from the displayed reset time and compare the cost of accepted coding work.
Last reviewed:
Read the dashboard or CLI status first
OpenAI directs users near a limit to Settings or the usage dashboard, where the exhausted allowance, credit balance and displayed reset time can be checked. In an active Codex CLI session, enter /status. There is no universal task allowance or reset timer on this page.
Check account identity, selected model, execution location and the active connection method. Keep a private note of the allowance name and timezone. Do not infer remaining capacity from a plan name alone, or apply another account’s reset screenshot to your work.
Identify which credential pays for the task
Classify the session as an authorized ChatGPT-account allowance or an independently billed API/provider credential. A gateway base URL, model ID and API key are not proof that a ChatGPT subscription pays for the requests. Read the tool configuration and the corresponding billing record.
For custom gateway configuration, follow the linked guide and current official provider reference. Use a named provider and a credential environment variable. Keep the original settings so they can be restored; this usage page does not certify any third-party endpoint or claim a paid request was tested.
Do not assume a fixed limit multiplier
The official guidance explains that workload depends on model, task complexity, context, reasoning, execution location and tools. Account allowances and credits can be shared with other eligible features. The current account dashboard is more reliable than a stale headline task count.
Current Pro tiers and some older Pro 200 allowance conditions differ. The reviewed official article says eligible grandfathered accounts retain the previous allowance through October 29, 2026; check that article and your account for the transition. A reset or an upgrade should never be promised from this editorial page.
A worked scheduling and cost example
Illustrative schedule: if the dashboard shows the relevant allowance resetting at 14:30, schedule a small verification after 14:30 and leave time for review before a 16:00 deadline. Do not begin a large task just because a different allowance reset. These times are planning examples, not a live account status.
Illustrative cost log: a $20 monthly subscription allocated across 10 accepted patches is $2 per accepted patch. If a separate assumed API bill of $8 supported those same patches, the combined allocation is ($20 + $8) ÷ 10 = $2.80. Count retries and rejected patches in the spending total; the task count and $8 are scenario inputs.
Keep evidence for the next capacity decision
Record date, model, local or cloud task, active credential category, accepted result, retries, review minutes, exhausted allowance, reset shown and additional charge. Use the same acceptance criteria during a normal week so a plan comparison has a consistent basis.
When work stops, preserve the non-sensitive error, account display and time. Check whether waiting, reducing repeated work or purchasing an allowed option addresses that specific limitation. Capacity and output quality are different problems, so compare both before changing the plan.