Home · Guides · Codex usage limits
Codex usage limits, explained
OpenAI's Codex CLI meters your ChatGPT plan through a 5-hour window, a weekly cap, and a credits system. Here's how the three interact, what the limit messages mean, and how to see your remaining quota.
Checked against official docs: July 13, 2026Naming note: this page is about the Codex coding agent (CLI, IDE extension and cloud tasks) that ships with ChatGPT plans — not the deprecated 2021 "Codex" code-completion model of the same name.
How the limits are structured
- 5-hour rolling window — shared between local CLI messages and cloud tasks. official
- Weekly cap — sits on top; resets 7 days after your first message of the week. official
- Credits as the unit — Codex counts usage in credits tied to token consumption, and the rate depends on the model. GPT-5.6 Sol costs roughly 5× the credits of the mid-tier model per token; a typical message runs 5–40 credits. Choosing a lighter model stretches the same window dramatically. official
- Jul 12–13, 2026 — after the GPT-5.6 Sol launch doubled traffic, OpenAI temporarily removed the 5-hour limit for Plus/Business/Pro; the weekly cap is currently the only ceiling. Announced by an OpenAI staff member as temporary, no end date given — it can return at any moment.
Limits by plan
| Plan | Price | Codex access | Notes |
|---|---|---|---|
| Free / Go | $0 | Yes, reduced | Lighter default model, smaller windows; no credit top-up — hitting the wall offers an upgrade, not a purchase |
| Plus | $20/mo | Full | 5-hour window + weekly cap; can buy extra credits and enable auto top-up |
| Pro (5x) | from $100/mo | Full, 5× Plus | Same structure, bigger buckets |
| Pro (20x) | higher tier | Full, 20× Plus | For sustained heavy agent use |
OpenAI publishes multipliers and credit rates rather than absolute hour counts; effective capacity depends on which model you run. The definitive numbers for your account are on chatgpt.com/codex/settings/usage.
What happens when you hit the limit
Real messages users see (wording varies slightly by plan and surface):
You've hit your usage limit. Upgrade to Pro or try again in 3 days 13 hours 6 minutes.You've hit your usage limit. Visit codex/settings/usage to purchase more credits or try again at 9:01 PM.- On Free/Go:
You've hit your usage limit. Upgrade to Plus to continue using Codex— lower tiers can't buy credits at all.
Plus and Pro can buy extra credits or switch on Auto top-up in Codex Settings → Usage. There are also documented mismatch bugs — reports of "usage limit reached" while the dashboard shows quota remaining — which is exactly why an independent eye on the numbers helps.
How to check your remaining quota
- In the CLI:
/statusduring a session — 5-hour and weekly remainders, model and plan, reset countdowns. - On the web:
chatgpt.com/codex/settings/usage— the official dashboard (it's the URL Codex's own error messages point to). - Status line: add
five-hour-limit/weekly-limittotui.status_linein~/.codex/config.toml. - Passively: keep it in the menu bar. AI Status Bar reads your existing local Codex login and shows both windows without you running anything.
Note: ~/.codex/auth.json holds your session credentials, not a local quota ledger — any tool showing live numbers is fetching them from OpenAI with your login, there is no offline source of truth.
FAQ
Do local CLI usage and cloud tasks share the same limit?
Yes — the 5-hour window covers both local messages and cloud tasks on your plan.
Does model choice change how fast I burn quota?
Dramatically. Credits per token differ roughly 5× between the top model and the mid-tier one — running a lighter model is the single biggest lever for stretching your window.
Is the 5-hour limit gone for good?
No — the July 2026 removal is explicitly temporary, a capacity response to the GPT-5.6 Sol launch rush. Plan for it to come back.
Codex says I hit the limit but my dashboard shows quota left. What gives?
Known reports of exactly this mismatch exist on OpenAI's tracker. Retry after a few minutes, check the web dashboard, and if it persists, it's a bug worth reporting — not your quota lying to you.