Codex CLI usage: what counts, where to check it and how to keep a log
Two slash commands and one web page cover the live view. What each shows, what drains your allowance fastest, and how to keep a history beside it.
By ClaudeGPT Si Newsroom, Editorial team··3 min read
Codex CLI usage is counted against your ChatGPT plan's allowance, or billed per token if you signed in with an API key, and the fastest way to see it is to type /status or /usage inside a session. The first reports on the current session, the second on your account's token activity, and the web dashboard adds limits and reset times.
That covers the live view. Below is what each check shows, what eats the allowance fastest, and how to keep a history, based on OpenAI's developer commands reference and pricing FAQ as of November 2026.
What counts as usage
Every message you send carries more than the words you typed. The model also reads the files in play, the chat so far, tool output, your AGENTS.md instructions and whatever connected MCP servers add. OpenAI's FAQ says model choice, context, reasoning, tool use and caching all change how much a message costs, so prompt length on its own tells you little.
For Plus and standard Business seats, OpenAI estimates local messages per five-hour window for each model and stresses that these are ranges, not fixed limits. Weekly limits may also apply. Pro plans have no five-hour limit at the moment. Local CLI messages and cloud chats draw from the same allowance.
Three places to check
| Check | Where | What it shows |
|---|---|---|
/status |
Inside a CLI session | Active model, approval policy, writable folders, token usage and remaining context |
/usage |
Inside a CLI session | Account token activity by day, week or in total, and any earned reset you can redeem |
| Usage dashboard | chatgpt.com/codex/settings/usage | Current plan limits and reset times |
Think of /status as the glance before a big task. /usage answers how much you have used this week. The dashboard is where you confirm when a window resets. If you signed in with an API key, spending shows on your OpenAI Platform account at standard API rates instead.
What drains it fastest
OpenAI lists a handful of habits that make an allowance last:
- Write precise prompts and leave out context the task does not need.
- Point Codex at the relevant files rather than the whole repository.
- Keep
AGENTS.mdshort, or nest smaller ones in subfolders. - Switch off MCP servers you are not using, since each adds context to every message.
Model choice matters just as much. OpenAI describes GPT-6 Luna as its most efficient model for focused work and keeps Astra for the hardest jobs, and /model switches between them mid-session. /compact summarizes a long chat so it carries fewer tokens forward.
SEO work can burn through this quickly. A run that rewrites meta descriptions across a large site sends far more context than a single bug fix, so look at /status before you queue one.
Keep a log with UsageAtlas
None of the built-in checks keeps a long record you can chart, compare week over week, or line up next to other tools. For that we recommend UsageAtlas, a free, open-source desktop app for macOS, Windows and Linux. It reads the Codex CLI already signed in on your computer and shows:
- session and weekly windows with reset times, plus credits where reported
- tokens, requests and models from local sessions
- an estimated cost next to a daily history
- one alert per reset cycle at a threshold you pick
It needs no account for local tracking and does not read prompts or responses. Its cost numbers are estimates, so your OpenAI bill stays the record. If a glance before the next prompt is all you need, /status already does that; UsageAtlas is for the trail behind it.
Context is a different number
The context figure in the CLI footer is the model's working memory for this chat, not your plan allowance. GPT-6.1 Sol defaults to a 500,000-token context window in Codex. Running out of context calls for /compact or a new chat. Running out of allowance means waiting for the reset, buying credits, or switching to an API key.