How to check your Codex usage limits

Codex is OpenAI’s coding agent, included in every ChatGPT plan from Free to Enterprise, and each plan meters it differently. Plus works in five-hour windows, with weekly limits that may also apply. Pro tiers currently have no five-hour window. Business and Enterprise can run on credits. In every case OpenAI shows the limit as a share of your allowance and a reset time, not a fixed number of messages. The model, the task and where it runs all change how much one request costs.

This guide shows where Codex reports those numbers in the CLI, the IDE extension and the usage dashboard, what happens when you run out, and OpenAI’s own advice for making an allowance last. Claude Code counts its limits differently; that is covered in how to check your Claude Code usage limits.

How Codex usage limits work

Local messages in the CLI, the IDE extension and the desktop app share one allowance with cloud tasks, and cloud tasks may use more of it. On plans that use shared allowances or credits, Codex also shares that pool with ChatGPT Work, ChatGPT for Excel and Workspace Agents.

Prices are from OpenAI’s Codex pricing page and About ChatGPT Pro tiers. OpenAI publishes estimates rather than fixed limits. For Plus and Standard Business it gives these ranges of local messages per five-hour window:

Speed settings count too. Fast mode uses included usage 2.5 times as quickly as Standard, and GPT-6 Astra Ultrafast (among Pro tiers, only on Pro 500) 8 times as quickly. Image generation uses included limits about 3 to 5 times faster on average. Among Pro tiers, OpenAI says Pro 200 includes more usage than Pro 100 and Pro 500 the most, without publishing the numbers.

How to check your Codex usage

These commands are from OpenAI’s slash command reference.

1. Type /status in the Codex CLI. OpenAI points here for your remaining limits during an active session. The same output confirms the model, approval policy, writable roots and current token usage.

2. Put the limits in the CLI footer. Type /statusline, turn on the rate limits item and confirm. The footer updates at once and the choice is saved to tui.status_line in config.toml, so the limits stay visible while you work.

3. Use /usage for token activity. /usage daily, /usage weekly or /usage cumulative shows your ChatGPT token activity inside the terminal. The same menu redeems an earned rate-limit reset when you have one.

4. Open the usage dashboard. chatgpt.com/codex/settings/usage shows your current limits and reset times. Your credit balance, and the place to buy more, is under Settings → Usage in ChatGPT or Usage & Billing in Codex.

5. In the IDE extension, type /status in the chat. It shows the chat ID, context usage and rate limits without leaving the editor.

OpenAI suggests checking the dashboard every week or two to learn your pace. If usage runs higher than expected, consider whether a smaller model or a tighter task would still do the job.

When does the Codex limit reset?

The usage dashboard tells you: chatgpt.com/codex/settings/usage shows your current limits and when each one resets, and /status shows what is left during a CLI session. How often a limit comes back depends on your plan, according to OpenAI’s Codex pricing page:

• Plus and Standard Business: local messages run in a five-hour window, and weekly limits may also apply.

• Pro: there is no five-hour limit at the moment. Other limits, such as weekly ones, can still apply.

• In the middle of a task: if you reach a limit during an active turn, Codex can finish that turn, subject to fair-use limits.

A reset that OpenAI gives you works differently from this schedule; see what happens when you hit a Codex limit.

What happens when you hit a Codex limit

OpenAI lets work in progress finish: if you reach a limit during an active turn, Codex can keep working on that turn, subject to fair-use limits. After that, the limit banner and the usage page list what your account can do, as OpenAI’s help article Using Codex with your ChatGPT plan explains.

• Buy credits. Plus and Pro users can buy credits and keep working without changing plan. Included usage is used first, then the credit balance.

• Use a reset, if you have one. OpenAI occasionally gives eligible users a one-time reset, and some Plus and Pro referral offers add a banked one. A full banked reset refreshes both the five-hour and the weekly window and moves your weekly reset date; Settings → Usage shows the new time.

• Run extra local chats with an API key. Any user can sign in to the CLI or IDE extension with an API key and pay standard API rates for the extra work. API-key sessions do not draw on your ChatGPT plan.

• Upgrade, or wait. The banner shows the reset time. Support cannot reset limits for you.

ChatGPT’s own limits are separate. Banners about images per day, voice caps or file uploads in ChatGPT do not apply to Codex.

How credits work

Credits are the unit OpenAI uses to pay for eligible usage on credit-based plans, and on Plus and Pro they let you keep working after your included limits. Each model has a rate in credits per million tokens, with input, cached input and output priced separately. For example, GPT-6.1 Sol costs 50 credits per million input tokens, 2.5 per million cached input tokens and 250 per million output tokens, while GPT-6 Luna costs 2.5, 0.25 and 12.5. Credit rates do not tell you how quickly you use included subscription limits, so OpenAI points back to the dashboard for that.

From building Orcylo: in the Codex logs on the Mac we build Orcylo on, about 98 % of input tokens were cached input. At GPT-6.1 Sol’s rates (2.5 credits per million cached tokens against 50 fresh), that mix puts about 70 % of the input cost on cached tokens, even though they cost a twentieth as much. With logs like these, the cached rate matters more than the headline one.

How to make Codex limits last longer

The first five come from the FAQ on OpenAI’s pricing page; the last two follow from its rate tables.

1. Keep prompts tight. Be precise about the task, and leave out context the agent does not need.
2. Limit source material. Give it only the relevant files, and narrow the sources or date range when you can.
3. Ask for the output you need. Say who it is for, the format and the length, and separate required work from nice-to-haves.
4. Shrink AGENTS.md. On large projects, nest AGENTS.md files in the folders they apply to instead of loading one big file for everything.
5. Turn off MCP servers you do not need. Each one adds context to every message.
6. Pick the model for the task. OpenAI’s estimates put GPT-6 Luna at hundreds to thousands of local messages per window, against 5 to 45 for GPT-6 Astra.
7. Leave Fast mode off for routine work. It uses included usage 2.5 times faster than Standard.

Where Codex keeps its records on your Mac

Codex saves session transcripts in its home folder, ~/.codex unless you set CODEX_HOME, with sessions under ~/.codex/sessions. They add up: a week of Codex logs on the Mac we build Orcylo on came to about 750 MB. OpenAI points to the history.persistence and history.max_bytes settings in config.toml, described in its advanced configuration guide, if you do not want Codex to keep transcripts. Depending on your settings, the same folder can also hold auth.json with your sign-in, so a tool that reads your usage from here only needs the session files, never that one.

What Orcylo adds

Orcylo is a menu bar system monitor for Mac that reads Codex’s session logs and shows your exact plan limits: the five-hour and weekly windows and your plan type. It works for Codex in the CLI, the IDE extension and the ChatGPT app. Each limit gets a ring and a countdown to its reset. Two bars compare how much of the window has passed with how much of the limit you have used, and turn orange when use runs ahead. Notifications come at 80 % and 95 %, when you reach a limit and when it resets, and a banner says why Codex stopped serving: rate, credits, workspace or spend limit.

Orcylo also shows your credits balance and your tokens by day, hour, model and project, with cached input counted separately. A Codex item can sit in the menu bar next to the Claude one and your CPU and memory. Orcylo never opens auth.json or any other credential file.

Orcylo costs €5 at launch for the first 20 licences, then €7.50 for the next 30, then €9.99, once, for one Mac (pricing). If you also want to compare free trackers, see the best Claude Code usage trackers for Mac; several of them cover Codex too.