Headroom · for Codex people
Your Codex usage,
on a fuel gauge.
One weekly pool, a 5-hour window on pause, credits that kick in only when the plan runs dry. Headroom reads what OpenAI reports for your account and keeps it in the Windows taskbar.
Windows 10 & 11 · signed auto-updates · no account needed
How Codex usage limits actually work
What counts, when it resets, and what happens at zero.
Codex usage is metered per ChatGPT account, not per tool. The CLI, the IDE extension, cloud tasks and code review all draw one shared pool on your sign-in. Burn scales with tokens and task complexity rather than prompt count — OpenAI no longer publishes fixed per-plan numbers, and long agent loops cost far more than short exchanges. API-key usage is the exception: it bills to your API organization and never touches these limits.
Two windows govern that pool on paper; one is live. The weekly window, roughly 7 days on your account's own cadence, is the limit that actually runs out. Its 5-hour sibling has sat unenforced since mid-July 2026, with no formal word on whether it returns. Hit zero and Codex refuses new tasks until the countdown ends — unless you bridge the gap with purchased credits or a banked reset.
The weekly limit
Roughly 7 days of shared pool for everything Codex does on your sign-in: CLI, IDE extension, cloud tasks, code review. There is no universal reset day — the window runs on your account's own cadence, and the countdown on OpenAI's usage dashboard is the only authoritative one. At zero, Codex turns down new work until the window rolls over; Plus and Pro users can buy credits, redeem a banked reset, or upgrade.
The 5-hour session window
Historically a rolling window that restored about 5 hours after the usage that consumed it, across the ChatGPT plans that include Codex. OpenAI stopped enforcing it in mid-July 2026 for Plus, Pro and Business; press called the pause temporary, but no written commitment exists and it could return without notice. While both windows ran, a 5-hour reset never restored weekly quota.
Codex credits
A purchased balance Codex draws down only after your plan's included usage is exhausted, never before. OpenAI's help center describes credits as valid for 12 months with no rollover past expiry, and eligible Plus and Pro accounts can turn on auto top-up. When plan and balance are both empty, Codex stops until a window resets or you buy more.
Banked rate-limit resets
Stored resets you trigger by hand when a limit empties, instead of waiting out the countdown. Third-party reporting on the June 2026 rollout says Go, Plus, Pro and Business start with one free reset, earn more through referrals, and get about 30 days to use each. Not the same thing as paid credits.
What Headroom shows for Codex
Exactly the windows OpenAI is enforcing right now, read from your account's own usage data with the CLI sign-in you already have — read-only, refresh token untouched.
The week as one number
While you work, the tray icon shows percent left on the window that bites first. With the 5-hour limit paused, that's your week: one number, amber when it's getting low, in your peripheral vision instead of a browser tab.
The pause, on the record
The 5-hour row stays visible as a "not enforced" ghost instead of quietly vanishing. If OpenAI switches the window back on, the live meter returns by itself — no update required. The tick on the Week bar marks where an even burn would put you.
Credits and resets, kept apart
Purchased credits get a row that hides itself when your account reports them unlimited. Banked resets keep a separate count. Run Codex on an API key instead and it's another ledger entirely — Headroom's OpenAI API gauge meters that as monthly org spend against a budget you set.
Asked every week
When does the Codex weekly limit reset?
There's no universal reset day. The roughly 7-day window runs on your account's own cadence (community reports describe it as rolling from first use), and the countdown at chatgpt.com/codex/settings/usage is the authoritative answer for your account. Headroom shows the same countdown next to its Week meter, so the answer lives in your taskbar instead of a settings page.
Is the Codex 5-hour limit gone?
OpenAI stopped enforcing the 5-hour window in mid-July 2026 for Plus, Pro and Business accounts. Press coverage called the pause temporary, but no written commitment exists either way, so it could return without notice. The weekly limit still applies — and when both windows ran, a 5-hour reset never restored weekly quota.
Why is my Codex weekly limit going down so fast?
Consumption is weighted by tokens and task complexity, not by how many prompts you type — one long agent loop can consume a large share of the week. Cloud tasks, code review, and CLI or IDE work all draw the same pool on your ChatGPT sign-in. Watching the pace mid-week beats reconstructing what happened afterwards. That's what a live meter is for.
How do I check my Codex usage?
OpenAI's official readouts are the usage dashboard at chatgpt.com/codex/settings/usage, the /status command inside the CLI, and an opt-in CLI statusline. All work; all require you to go and look. Headroom reads the same account usage data and keeps it in the Windows taskbar continuously — it needs the Codex CLI signed in with ChatGPT, because API-key-only auth has no plan meters to read.
Do Codex credits expire?
They don't last forever: OpenAI's help center describes purchased credits as valid for 12 months, with no rollover past that date. Credits are drawn down only after your plan's included usage runs out, and eligible Plus and Pro users can enable auto top-up. Headroom shows the balance as a Credits row and hides it if your account reports credits as unlimited.
What are Codex banked rate-limit resets?
A banked reset is a stored voucher you redeem manually to refresh a limit you've already hit. Third-party reporting dates the rollout to June 2026 for Go, Plus, Pro and Business: one free to start, more earnable through referrals, roughly 30 days to use each. They're separate from paid credits, and Headroom counts them as their own Resets row.
Codex isn't the only plan with opinions. Headroom watches the rest, too: