Checking Codex Usage with /status and /usage

Codex gives you two different readings of what you're consuming, and they answer different questions. /status describes the session in front of you: which model is running, how it's configured, how many tokens it has used and how much context it has left. /usage describes your account: how much ChatGPT token activity you've used over a day, a week or overall. Budgeting well means reading both, at the right moments, and knowing what neither one tells you.

/status: the session in front of you

/status shows the session's configuration and token usage — the active model, the approval policy, the writable roots, and the remaining context capacity. Each of those has a cost angle. The model is the single biggest factor in what a task costs, and the default changes over time; the CLI now runs GPT-6.1 Sol unless something overrides it, so /status is the place to confirm what you're actually paying for. Remaining context tells you how close a long session is to its ceiling, and how much it re-sends every turn. And the token figures are the raw material for every estimate on this site.

Run it at the start of any session where cost matters, after /model or /fast, and before handing the session a long autonomous task. If you'd rather not keep asking, /statusline puts model, context, limits and token counts in the footer permanently.

/usage: the account behind it

/usage shows your ChatGPT token activity — daily, weekly or cumulative — from inside the terminal, and it's also where you use a rate-limit reset if your plan has one available. Plus and Pro users gained banked rate-limit resets earlier this year, so it's worth knowing the command exists before you hit a limit rather than after. For current limits and reset times, OpenAI points to the usage dashboard; the CLI view is for activity.

What drives the numbers

OpenAI is explicit that prompt length alone isn't a reliable estimate of usage. Model choice, context, reasoning, tool use, retrieval and caching all affect it, and tasks that look similar can consume very different amounts. A small script may use a fraction of an allowance while a long session that holds a large context uses far more per message. Speed modes add a multiplier on top: in a ChatGPT plan, Fast consumes included usage at 2.5x the Standard rate, and GPT-6 Astra Ultrafast at 8x.

Usage is also shared more widely than people expect. Local messages and cloud chats draw on the same allowance, and cloud tasks may use more of it than local ones. Usage limits are shared with other agentic features once their pricing takes effect — OpenAI currently names ChatGPT for Excel on Plus and Pro. So a quiet day in the CLI doesn't guarantee headroom if the same account is busy elsewhere.

Turning readings into a budget

The useful habit is to measure a few representative tasks rather than guess. Note the token usage /status reports at the end of each one, split as best you can into uncached input, cached input and output. Then:

• On a ChatGPT plan, put those figures into the Codex credits calculator. It prices the task in credits for any model and speed, and shows OpenAI's estimate of local messages per 5-hour window on Plus for scale. Plus and standard Business seats work in windows of that length, with weekly limits that may also apply; Pro currently has no window limit.
• With an API key, use the token cost estimator instead — API billing is per token at the model's published rates, and the usage tracker keeps a running total from real usage figures.

A handful of measured tasks beats any rule of thumb, because the spread between a small task and a large one is wider than the difference between most models.

When the numbers look too big

If /status reports far more tokens than the visible conversation seems to justify, the difference is usually context the session re-sends on every turn — file contents, tool output and earlier messages — plus reasoning, which is billed as output but never appears in the transcript. OpenAI's own pricing notes make the same point from the other side: extended sessions that hold more context use significantly more per message. That's the gap between "I only asked a few questions" and the usage you see, and it's why a long session costs more per message than a fresh one doing the same work.

Reading the trend, not just the snapshot

A single /usage check tells you where you are; the trend tells you whether a habit is costing you. If weekly activity jumps without a change in the work, look for the usual causes: a session left in Fast mode, a model change after a CLI update, long sessions that never get compacted, or cloud tasks running alongside local work. /compact summarizes a long chat to free tokens, which reduces what every later turn re-sends.

What these commands don't show

Neither command shows money. On a ChatGPT plan, what a credit costs depends on your plan or agreement, and this site never assumes a figure. On an API key, billing lives in the platform's Usage and Costs dashboards, which can filter and group by API key, and where you can set spend alerts or hard spend limits. Setting a soft budget alert covers that side, including when a hard cap is worth having.

A short routine

Check /status when a session starts and before a long autonomous run. Check /usage at the end of the day or week, not mid-task. Measure a new kind of task once and keep the numbers. And when a limit arrives sooner than expected, rule out Fast mode and the active model before blaming the work itself — Codex plans and usage limits shows how each plan meters usage.