To see your Codex token usage, type /status in the Codex CLI for the current session, or /usage for daily, weekly and cumulative token activity on your account. If you want that number kept by day, model and project after the session ends, UsageAtlas reads your local Codex activity and turns it into a history you can reopen.
Those are three different views of the same tokens. Which one you need depends on whether you are asking about right now, about your account, or about last Tuesday.
Where Codex shows token usage #
OpenAI's Codex CLI has two built-in commands for this, plus a web page.
| Where | What it shows | Best for |
|---|---|---|
| /status | Session configuration, token usage and remaining context capacity | Checking the chat you are in |
| /usage daily, weekly or cumulative | Account token activity over that period | A quick total without leaving the terminal |
| Usage dashboard on chatgpt.com | Plan limits and reset times | Checking how much of your plan is left |
| UsageAtlas | Tokens, requests, models and estimated cost by day, with history | Looking back, comparing days, splitting by project |
The footer in the Codex terminal UI also shows how much context is left, and /statusline lets you add token and limit fields to it.
Step by step: see Codex token usage on your desktop #
- Install UsageAtlas on the machine where you run Codex. It is free and open source, and it runs on Windows, macOS and Linux.
- Sign in to Codex as usual. UsageAtlas reads the signed-in Codex CLI and the session activity already on your computer. No UsageAtlas account is needed.
- Open a day or a date range. You see tokens, requests, the models used and an estimated cost.
- If you also use Claude Code, Cursor or other tools, filter to Codex so their tokens drop out of the total.
- Open History and group by project to find the repository that used the most.
UsageAtlas does not read your prompts or responses. It works from usage figures, not from what you wrote.
Reading the token mix #
A single total hides a lot. Codex counts input tokens, cached input tokens and output tokens separately, and OpenAI prices them at different credit rates. An agent resends its conversation on each turn, so a long session can show a very large number that is mostly cached input.
UsageAtlas puts the input, output and cache split next to the headline figure. A day that looks huge in tokens can turn out modest once you see that most of it was cache.
What a heavy day usually means #
When one day stands out, check three things in order. First the model: OpenAI's Codex rate table prices models very differently per token, so a switch to a larger model shows up quickly in estimated cost. Then the project, since one repository with long test output can carry the day. Then the split, to see whether the extra came from long chats or from a lot of new output.
Tokens are not credits or limits #
Token usage tells you what Codex did. Your plan limit tells you what is left. Credits are what you spend after the included limits run out, on plans that offer them. OpenAI says its API token prices are separate from subscription usage, so do not convert one into the other.
UsageAtlas shows Codex session and weekly windows and credits where Codex reports them, beside the token history. Its cost figures are estimates from token counts and known model prices. Your OpenAI account is the record for billing.
When the built-in commands are enough #
If you only care about the chat in front of you, /status covers it, and /usage gives you a quick account total. A desktop history pays off when you want to compare weeks, see which model or project drove a heavy day, or keep Codex and Claude Code in one view.
[Download UsageAtlas](https://usageatlas.com) to keep your Codex token history in one place.