ChatGPT Manual

Codex goal

Codex goal means giving Codex a result to pursue. You still review the diff. The work spends the same Codex allowance as other tasks.

ChatGPT logo
The mark is the ChatGPT logo from OpenAI. This manual does not sign you in.

A goal is an outcome, not a vibe

Codex goal searches land on OpenAI's follow-goals use case. The idea is to state a finish line the agent can check: tests passing, a migration applied, a bug no longer reproducing. 'Make it better' is not a goal. 'The login form rejects an empty email and the existing test covers it' is.

Goals spend Codex usage. A long chase on a large model can hit the five-hour or weekly window before the goal is met. Ask for a plan first if the repo is unfamiliar, then let it run a smaller slice.

What you still review

Read the diff. A goal can be marked done because a test was deleted or a check was weakened. Tell Codex which tests are not allowed to change. If the work needs a second agent, that is the subagent topic, and it spends more of the same allowance.

This manual cannot see your repository and does not run Codex. Sign in with the ChatGPT account that owns the plan.

Questions

Do goals cost extra?
They use the Codex allowance included in your plan, and credits if you have bought them. There is not a separate goal fee listed on the pricing page.
Can I leave a goal running overnight?
Only if your limits and the product's run rules allow it. Check the usage panel before you walk away, and expect to review the result in the morning.
Is a goal the same as a custom GPT?
No. A goal is a Codex task with an outcome. A custom GPT is a saved chat configuration in ChatGPT.

Related guides

Sources

Checked 2026-09-25. Prices, limits, and model names can change on the live pages.