How to Track OpenAI Codex Token Usage

Published 16 May 2026

OpenAI Codex shows you how much of your subscription you have used. What it does not show is how many tokens that actually is.

"70 % consumed" could mean 300,000 tokens or 3,000,000 tokens. The difference matters when you are budgeting a multi-file refactor or comparing Codex against Claude Code on the same task.

This guide shows how TokenBBQ turns Codex's percentage into real, absolute numbers you can act on.

1. What Codex Shows You Today

Codex has a built-in status command you can run inside the CLI:

/status — shows your current subscription tier usage as a percentage

This is useful for a quick check, but it has limits:

It shows only a percentage, not absolute token counts
It does not show input versus output tokens separately
It does not track usage across multiple Codex sessions or other AI tools

The OpenAI Usage Dashboard gives you account-level totals, but it updates with a delay and does not split by tool or project.

2. How Codex Differs from Claude Code

Codex and Claude Code have different token profiles, which affects what you see in TokenBBQ:

Codex often reads selected context through tool calls instead of loading every file at once. The exact token profile depends on the prompt, model, and run mode.
Codex can chain tool calls, so long agent runs may accumulate input, output, cached input, and reasoning tokens quickly.
Codex exposes run controls such as --model, --ask-for-approval, --sandbox, and reasoning effort settings, depending on the installed CLI version.

Cost note: exact cost depends on the model name, cached input, reasoning tokens, and the current pricing table. TokenBBQ fetches LiteLLM pricing when available and leaves unknown model prices at $0 rather than guessing.

3. What TokenBBQ Adds

TokenBBQ reads the logs that Codex already writes to your machine and turns them into a dashboard. No proxy, no network interception, no code changes.

What you get	Codex native	With TokenBBQ
Absolute token count	No — percentages only	Yes — input + output
Cross-session daily total	No	Yes — auto-aggregated
Multi-tool comparison	No — only Codex	Yes — Codex, Claude, Gemini, etc.
Live dashboard	No	Yes — browser + CLI
Project-level split	No	Yes — by working directory

4. How TokenBBQ Works

TokenBBQ reads the log files that Codex writes locally. It extracts token counts, timestamps, and model information, then serves them in two forms:

Web dashboard — open in your browser, auto-refreshes every 5 seconds
Terminal output — npx tokenbbq daily, npx tokenbbq monthly, npx tokenbbq summary, or --json

Everything stays on your machine. The only network request TokenBBQ makes is to fetch current API pricing from LiteLLM so cost estimates stay accurate.

5. Quick Start

No install, no config, no API key:

npx tokenbbq@latest

This starts the web dashboard on http://localhost:3000. Run Codex as usual. TokenBBQ picks up the log data automatically.

TokenBBQ dashboard showing Codex token usage with absolute numbers

TokenBBQ dashboard — Codex token usage in absolute numbers

For terminal-only output:

npx tokenbbq@latest daily
npx tokenbbq@latest monthly
npx tokenbbq@latest summary --json

6. About the Desktop Pill

TokenBBQ also offers a separate desktop widget — the Pill — that floats above your IDE and shows subscription usage at a glance.

Codex subscription fill (Primary and Secondary window with reset time)
Tokens consumed today from all local AI tools
Toggle to switch the Pill between Codex and Claude Code view

The Pill is a separate desktop app with its own installer. It is not part of npx tokenbbq. One click on the Pill opens the full dashboard.

TokenBBQ desktop pill showing Codex subscription usage

The Pill — subscription fill and token count at a glance

7. What the Numbers Mean

TokenBBQ shows three metrics for Codex:

Input tokens — your code, prompts, and context sent to Codex
Output tokens — Codex's responses, diffs, and explanations
Estimated cost — based on OpenAI's published pricing per model tier

The cost estimate is not official billing. It is a quick sanity check based on current API rates.

8. Comparing Codex and Claude Code

TokenBBQ tracks both, so you can compare:

OpenAI Codex — selected-context, tool-driven runs can be efficient for quick refactors, and controls such as --model, --ask-for-approval, and --sandbox let you tune behavior.
Claude Code — often explores broader project context; TokenBBQ shows the actual usage instead of guessing from the tool name.

If you are hitting Codex rate limits daily, switching to Claude Code for deep reasoning tasks can keep you productive without upgrading your OpenAI tier. Read the Claude Code token tracking guide for the full setup.

9. Related Guides

AI Token Tracker — Compare Claude, Codex, Gemini & More — multi-tool comparison in one unified dashboard.
Track Codex Tokens with Absolute Numbers — convert Codex percentages into real token counts.
How to Track Claude Code Token Usage — the same setup for the Anthropic side, with the 5-hour and 7-day windows.
Track AI Token Usage Per Project — split daily Codex costs across multiple repositories or clients.

10. Summary

Codex's native /status command is fine for a quick look at your subscription fill. When you need absolute numbers, cross-session totals, or a view across multiple AI tools, TokenBBQ fills the gap.

Open-source, local-only, no sign-up. Run it once alongside Codex and stop guessing what "70 %" actually means.

Start tracking Codex tokens now

Open-source · MIT · Local-only · No sign-up

Get TokenBBQ →