Checking Codex tokens left starts with a quirk that surprises almost every new user: OpenAI never gives you a literal token balance. Instead, you have four separate views (in the terminal, in the browser and on your desktop), and ranking them by what they actually measure stops you from misreading a session number as an account cap.

Ranking the four ways to see remaining capacity

Rank Source What it shows Best used for
1 UsageAtlas Today Remaining percentage, reset countdown and local token history Planning work across resets and tools
2 Codex usage panel Official short and weekly window percentages Checking authoritative account status
3 /status inside Codex TUI Live context usage and current session tokens Spotting when one thread got too big
4 Provider billing page Credit balance and invoices Tracking paid overage or API spend

UsageAtlas ranks first for day-to-day workflow because it puts the official percentage and reset countdown right beside actual local token consumption from your session files, without waiting for you to open a browser tab.

Why no tool prints an exact token balance

There is no public OpenAI endpoint that returns a remaining token count for Codex subscription windows. Allowances operate in usage units across different models and cached input rates, so the stable metric exposed to accounts is a percentage.

To turn that percentage into a rough token figure for your current model:

  1. Note your starting percentage before a work block.
  2. Run your session and read tokens used from /status or UsageAtlas.
  3. Note the ending percentage and divide tokens used by percentage points lost.
  4. Multiply that per-point figure by your remaining percentage.

If 12,000 tokens lowered your window from 80% to 76% (4 points), one point is about 3,000 tokens and 76 points is roughly 228,000 tokens on that model. Switching models or hitting heavy cached input shifts the ratio immediately.

Two traps when reading remaining numbers

First, do not confuse context left in the current thread with account allowance left. Opening a fresh thread resets context to zero and restores room for files, yet leaves your five-hour and weekly window percentages untouched.

Second, watch both windows together. A five-hour window can bounce back after a break while the weekly budget continues tightening underneath it. Setting a low-capacity warning around 20% keeps you from running those calculations under deadline pressure.

Local tracking without sharing code

UsageAtlas is free, open source and local-first, needing no account for local tracking and never reading your prompts or responses. Cost numbers shown are estimates rather than official billing records, while OpenAI's own panel remains the final word on account limits.

FAQ

Can I find my exact Codex tokens left via an API?

No public OpenAI endpoint exposes an absolute token balance for Codex plan windows. Tools can only pair the reported percentage with observed local token totals.

Does /status show my weekly Codex allowance?

/status reports live context usage and token totals for the active CLI session, whereas the Codex usage panel and UsageAtlas show the short and weekly window percentages.

Why did my estimate change after switching models?

Different models carry different weights inside OpenAI's usage units, so the number of tokens per percentage point shifts whenever you change models.