/status in Codex tells you where you stand right now. It does not tell you which days you ran close to the limit, what used your tokens or which model you leaned on. The lower half of the Me view in Usage Insights for Codex does, from the numbers Codex already wrote into your session files.
Read your usage limits
Usage limits draws the highest share of each window you used, day by day: the 5-hour window and the weekly window. Under the chart:
- Peak 5-hour and Peak weekly: the most of each window you used in the range.
- Limit reached: how many times a window was at 100%, with your plan.
Codex records the limits only when you sign in with ChatGPT. With an API key, the panel says there are none in your data.
See your tokens
Tokens per day draws all your tokens for each day. Under it, Reasoning tokens says how many output tokens the models spent thinking before they answered, and, when subagents ran, Subagent share how much went to them.
Tokens & reasoning splits all of them, and its ring shows Cache efficiency: of all input, the share that was cached. Higher is faster and uses less of your limits.
Session files hold tokens, not prices, so there is no cost here. Events exported from an API-key gateway can carry a cost, and then the panel shows it.
Compare models and effort
Surface mix shows where your sessions ran (the terminal, the IDE extension, the app or codex exec), and Model mix under it which models answered. The chips under that count the responses at each reasoning effort. If most of your work runs at high, a lower effort for routine jobs leaves more of your limits for the hard ones.
Check your sandbox and approvals
Sandbox & approvals shows how hands-off you ran Codex: how many sessions were read-only, wrote only in the workspace or had full access, and which approval policy they used. A session with full access and no prompts is worth a second look.
Next: every skill, MCP server and tool Codex used, in the Capabilities view.