How to check your Codex usage limits

In the Codex CLI, type /status: OpenAI says it shows your remaining limits during a session. For the full picture with reset times, open the usage dashboard at chatgpt.com/codex/settings/usage. On Plus and Business the allowance is counted per five hours, and weekly limits may also apply. The CLI also writes the last numbers it got into its session files, where a script can read them.

Last updated: October 2026

Written by the developer of Crest, 8 October 2026. OpenAI’s pages are listed under Sources. We have no OpenAI account on our test Mac, so we ran Codex CLI 0.160.1 on a 16-inch MacBook Pro (Mac16,7) on macOS 27.0.1 against a small local server standing in for OpenAI’s, which sent limit numbers we chose: the 42% and 17% below are ours, not a real account’s. What Codex did with them is Codex’s own. Crest comes last, and you do not need it.

Where to look

The dashboard is the one to trust for reset times: OpenAI’s pricing page says to “Check your usage dashboard for current limits and reset times”, and suggests looking every week or two to understand your pace.

What the limits are

OpenAI counts Codex use per five-hour period on Plus and Standard Business, and adds that “Weekly limits may also apply.” Local messages and cloud chats share the same allowance, so a cloud task draws on the same limit as your terminal. How many messages fit depends on the plan and the model, and OpenAI keeps those numbers on its pricing page rather than in the CLI; check there, because they change. With an API key instead of a ChatGPT plan you pay per token and there is no plan limit to watch.

Comparing with Claude Code? Its limits work the same way in outline, a five-hour window and a weekly one, and how to check Claude Code usage limits covers its commands.

Put them in Codex’s status line

Codex’s footer can show your limits all the time. Type /statusline to pick the items, or set them in ~/.codex/config.toml:

# ~/.codex/config.toml
[tui]
status_line = ["model-with-reasoning", "five-hour-limit", "weekly-limit"]

Codex CLI 0.160.1 took this list without a warning. In our run the two limit items stayed empty, as /status said “data not available yet”; most likely because that Codex was not signed in to ChatGPT, since the session file did get the numbers. So we cannot show you what they print on a real account, only that these are the names 0.160.1 knows.

Where the CLI writes the numbers

Every Codex session writes a file under ~/.codex/sessions/, in folders by date, named rollout-<time>-<id>.jsonl. After each turn Codex adds a token_count line, and its rate_limits part holds what the server said about your limits. In our run it read:

{"limit_id":"codex","primary":{"used_percent":42.0,"window_minutes":300,"resets_at":1791495755},
 "secondary":{"used_percent":17.0,"window_minutes":10080,"resets_at":1791833135}, ...}

primary and secondary are the two windows; window_minutes says which is which (300 is five hours, 10080 is seven days), used_percent is the share used and resets_at is when it resets, in Unix seconds. The fields are defined in Codex’s open source code (protocol.rs). The interactive session wrote them too, even while its /status said no data.

This script prints them from your newest session:

#!/bin/sh
# codex-limits.sh: Codex's last-known limits, from its newest session file.
f=$(ls -t "${CODEX_HOME:-$HOME/.codex}"/sessions/*/*/*/rollout-*.jsonl 2>/dev/null | head -1)
[ -n "$f" ] || { echo "No Codex session yet"; exit 0; }
grep '"token_count"' "$f" | tail -1 | jq -r '
  .payload.rate_limits // empty
  | [.primary, .secondary][] | select(. != null)
  | "\(if .window_minutes == 300 then "5h" elif .window_minutes == 10080 then "weekly" else "\(.window_minutes) min" end): \(.used_percent | round)% used, resets \(.resets_at | localtime | strftime("%a %H:%M"))"'

Against our run it printed:

5h: 42% used, resets Fri 00:42
weekly: 17% used, resets Mon 22:25

Two limits to know. These are the last numbers from your last turn on this Mac, so work on another computer or in the cloud shows up only after your next turn here. And a session file holds your whole conversation; the script reads only the token_count lines, but do not paste those files anywhere you would not paste your code.

Or on the notch, next to Claude’s

Crest, a Mac notch app, reads the same rate_limits from your newest Codex session and shows “Codex limits” on its Agents screen, under Claude Code’s: the 5 hours share with its reset time and the 7 days share, in red from 85%. It reads only the end of that one file, never a credential, and places each window by its length, so a window it does not recognise is left out rather than shown under the wrong name. A reading more than 30 minutes old is not shown at all.

• A warning before you run out, on the closed notch: when the 7-day limit crosses 85%, or when the 5-hour window would run out before it resets at your recent pace. Once per window; Settings > Coding agents > “Warn before limits run out”, on by default.
• What is free: answering Codex approvals from the notch is free, with no account (Codex approvals in the notch). The limits and the warnings are part of Crest Pro.

1. Download Crest. Signed and notarized.Free tier, no account · 7-day Pro trial, no card · macOS 14 or later
2. Open Crest’s Settings, go to Coding agents, and keep Codex switched on in the list of agents Crest follows.
3. Run a Codex turn; the limits show after it.

Crest Pro, with Take and Minutes, is $24.99 once or $2.99 a month, and signing in gives you 7 days of it free.

Sources

• OpenAI, Codex docs, Pricing (developers.openai.com/codex/pricing redirects there). Read 8 October 2026. Messages “per five-hour period for Plus and Standard Business”, “Weekly limits may also apply”, the usage dashboard, /status during a CLI session, and “Check your usage dashboard for current limits and reset times”.
• OpenAI, Codex docs, Slash commands and the configuration reference. Read 8 October 2026. /status shows “the chat ID, context usage, and rate limits”; /usage opens usage and billing settings; tui.status_line is an ordered list of footer items.
• OpenAI, codex-rs protocol.rs on GitHub. Read 8 October 2026. TokenCountEvent.rate_limits, a RateLimitSnapshot with primary and secondary windows of used_percent, window_minutes and resets_at (Unix seconds).
• Our own runs, 8 October 2026, on a 16-inch MacBook Pro with macOS 27.0.1 and Codex CLI 0.160.1, in a throwaway CODEX_HOME. We have no OpenAI account on this Mac, so a small local server played OpenAI’s and sent limit headers we chose (42% and 17%); the session file, /status, the status line and the script above are what Codex and the script did with them.
• Crest source, read 8 October 2026: CodexLimits.swift, UsageBridge.swift, AgentSessionsReading.swift, QuotaWatch.swift, and the 6.1.0 entry of Changelog.swift.

FAQ

Keep reading: Claude Code usage limits · Copilot CLI notifications · Claude Code notifications · Codex notifications · Claude Code vs Codex · Crest overview