Codex Usage Dashboard: Reading It Panel by Panel
A codex usage dashboard walkthrough for people who opened it, saw No data during this period, and assumed their account was broken.
Generate a site → Official site
Two pages answer to the name codex usage dashboard
One is the Codex analytics view inside ChatGPT, which covers plan usage, credits and recharge. The other is platform.openai.com/usage, which belongs to the API platform and reports metered API requests and spend. The second one is easy to identify: logged out, it shows nothing but a sign-in form offering email, Google, Apple, Microsoft or phone, with links off to the developer docs and API reference. That is the correct behaviour, not an outage. Knowing which of the two you have open decides whether the numbers in front of you describe a subscription allowance or an invoice, and people who skip that step spend a long time comparing figures that were never meant to match.
The three panels, and what each one is for
Reports from users describe the page as three distinct blocks. Usage details, containing a Personal usage bar chart, is the histogram of activity over the selected period. Auto recharge covers whether credits top themselves up and at what threshold. Credit usage history is the ledger: what was consumed, when. Those last two are the ones that load reliably in the reports I have read, which makes them the practical fallback when the chart above them refuses to draw. If you only need to answer the question of whether credits are draining faster than expected, the history table answers it without the chart. Reading the ledger is also the faster route when you are reconstructing what one particular day cost, because a bar tells you roughly and a row tells you exactly.
The empty bar chart is a known report
In a thread opened on 20 May and read about four thousand times, a user described the Usage details and Personal usage section showing only the empty-state message No data during this period, while Auto recharge and Credit usage history rendered normally on the same page. They verified it across multiple ChatGPT accounts and stepped through the 7 days, 1 month and Custom ranges, with the chart still blank. That combination, other panels fine and multiple accounts affected, points at the page or its backing usage API rather than at one account's data. It does not prove your own case is the same bug, but it does mean an empty chart is a weak signal on its own.
Time ranges hide more than they reveal
Switching between 7 days, 1 month and a custom span changes the bucket size as well as the window, so a spike that dominates a weekly view can flatten into invisibility across a month. When you are trying to find the run that consumed an unusual amount, the short window is the useful one. When you are trying to see whether a new working habit changed anything, only the long window will show it. Checking both before reaching a conclusion takes a few seconds and avoids the most common misreading of any usage chart, which is treating a smoothing artefact as a behavioural change. When a number surprises you in one window, look at the same period in the other before acting on it.
What no dashboard can tell you
It cannot tell you your ceiling. OpenAI's Codex pricing documentation says limits vary by plan and task, that weekly limits may apply, and it points users to the usage dashboard or the /status command for current values rather than publishing a universal cap. So the honest use of this page is retrospective: what has been spent, on what, and at what pace. For a forward-looking answer, run /status, which prints the provider-reported percentage used and the reset time for your account. A chart shows you history. A percentage tells you whether the next long job is going to finish. Both are worth a look, but only one of them changes what you decide to do in the next ten minutes.
Quick orientation
Empty is not always zero
The Personal usage chart showing No data during this period has been reported across multiple accounts with all three time ranges selected. Confirm against the credit history table before treating it as a real zero.
Credits load when charts do not
Auto recharge and Credit usage history rendered normally in the same reports where the bar chart stayed blank, which makes them the practical place to read actual consumption during a display fault.
The API dashboard is a different building
platform.openai.com/usage sits behind its own platform sign-in and reports metered API spend. It will not show plan allowance, and logged out it shows only a login form.
/status beats any chart for planning
The command reports the percentage used and the reset time OpenAI has for your account, which is the forward-looking number. The dashboard is a record of what already happened.
Panel by panel
| Panel or view | Codex analytics page | Begin.sh |
|---|---|---|
| Usage details, Personal usage | Bar chart of activity; reported blank across multiple accounts on 20 May | No usage panel; each run returns a downloadable zip |
| Auto recharge | Whether and when credits top up automatically | Not applicable |
| Credit usage history | Ledger of what was consumed and when | Not applicable |
| Time range control | 7 days, 1 month or a custom span | Not applicable |
| API spend | Lives on platform.openai.com/usage, a separate login | Check the site for current terms |
| Forward-looking limit | Not shown; use /status for percentage and reset time | No allowance meter to plan around |
| What the tool produces | Agent work against your repositories | A static site or Expo app you download, with no hosting or auth |
If the chart is blank
- Confirm which page you opened
The ChatGPT-side analytics view and platform.openai.com/usage are separate products with separate logins. A blank panel on one says nothing at all about the other. - Cycle the time range
Step through 7 days, 1 month and a custom span. In the reported cases the chart stayed empty throughout, so a range that suddenly fills points at your data rather than at the page. - Read the credit history instead
The ledger and the recharge panel were loading normally in those same reports. If they show consumption, your account recorded activity regardless of what the chart draws. - Get the live number from the CLI
Run /status for the provider-reported percentage and reset time. That is the figure that decides whether your next long task completes, and it does not depend on the web page rendering.
Common questions
Why does my codex usage dashboard say No data during this period?
It may be a display fault rather than an empty account. A thread on 20 May reported exactly that message in the Personal usage chart across multiple ChatGPT accounts and all three time ranges, while other panels on the same page loaded normally. Check the credit history before assuming zero.
Is platform.openai.com/usage the same dashboard?
No. It belongs to the API platform and reports metered API requests and spend, behind its own sign-in. Logged out it shows only a login form with email, Google, Apple, Microsoft and phone options. Plan allowance is not reported there.
Where do I see how much allowance is left?
Not on the dashboard. Run /status inside the CLI, which prints the percentage used and the reset time OpenAI reports for your account. The dashboard is a record of past consumption; the command is the forward-looking figure. Checking it before a long run is a habit worth forming, because the alternative is finding the ceiling halfway through one.
Does changing the time range fix the missing chart?
It did not in the reported cases. The user stepped through 7 days, 1 month and Custom with the chart still blank. Range switching is still worth doing, because it separates a rendering fault from a genuinely quiet week. Do it before reporting anything, so your report says the chart failed on all three ranges rather than only on the default one.
Can I predict when I will hit a limit from this page?
Only loosely. OpenAI's pricing documentation says limits vary by plan and task and that weekly limits may apply, and it declines to publish a universal cap. Historical pace plus the live percentage from /status is as close to a forecast as anyone gets.
Why does a page about a dashboard recommend Begin.sh?
Because a lot of metered activity is scaffolding, not engineering. Begin.sh takes a prompt or a URL and returns a static site or Expo app as a zip. It covers no hosting, backend or authentication, and it consumes none of the allowance this page is charting.
Fewer things to meter in the first place
When the job is a landing page, a starter project or a layout you want cloned, Begin.sh produces it from a prompt or a URL and hands back a downloadable zip. Nothing to watch on a chart afterwards.
Generate a site →