Direct answer
Enter an all-time AI coding token total. The calculator counts eligible totals in the current public leaderboard that are less than or equal to yours, then divides by the eligible sample size. The result is a cohort position—not a score for engineering quality, productivity, model performance, market share, or plan allowance.
The comparison uses 231 eligible public rows from the all-time leaderboard snapshot generated Aug 28, 2026, 9:56 AM UTC.
Usage percentile
6.1th
14 of 231 eligible totals are at or below the entered value.
Entered all-time total
100,000,000
Tokens across the supported usage represented by your chosen total.
Current cohort median
4,333,380,044
Nearest-rank 50th-percentile threshold.
Current cohort p90
28,244,747,391
Nearest-rank 90th-percentile threshold.
Cohort position
0th percentile → 100th percentile
At or below
14 / 231
Median threshold
4,333,380,044
P90 threshold
28,244,747,391
Transparent method
The server requests the public leaderboard's all-time token view and removes missing, non-finite, zero, and negative totals. It passes only numeric totals to the browser calculator—never handles, display names, avatars, or social identities. Your entered value stays in the browser. The calculator includes ties at your value in the below-or-equal count, so identical totals receive the same percentile under this rule.
The displayed median and p90 use the nearest-rank method. After sorting eligible totals from low to high, the median is the value at ceiling(n × 0.50), and p90 is the value at ceiling(n × 0.90), with one-based ranks. With a small sample, thresholds can move sharply when one participant appears, opts out, or updates an all-time total.
Interpretation guardrails
Compare an all-time total with this calculator. A daily, weekly, monthly, or single-session number has a different exposure window and will create a misleading position.
If several listed developers have the same total as the entered value, all of those rows are included. This makes the rule reproducible but can create visible percentile jumps.
It excludes developers who never participate, do not publish, opt out, or use sources that are not collected. It is not a representative survey of developers or AI-tool users.
A higher total may reflect a longer history, larger contexts, more cache reads, more agents, heavier work, inefficient retries, or other measurement differences. It does not establish quality or productivity.
Go deeper
This tool answers one personal comparison question. The benchmark report explains the broader distribution, sample coverage, quartile thresholds, and limitations without turning a single percentile into a performance judgment.
Read the live token usage benchmarksThe input sample is the current public all-time leaderboard response sorted by tokens. Only finite positive totals are eligible. The live sample can change as listed participants sync new history, join, hide their profile, or are removed, so a saved percentile should always retain the displayed generation time and sample size.
This is an empirical rank within one opt-in product cohort. It is not a provider quota, subscription allowance, billing reconciliation, industry market-share statistic, model benchmark, developer ranking, or measure of code shipped. Differences in account age, tool coverage, model tokenization, cache accounting, and working style all affect totals.
The nearest-rank median and p90 are observations from the sorted sample, not interpolated predictions. Ties are preserved and included in the at-or-below numerator. No percentile is shown when the source is unavailable, intentionally redacted, or empty; the page does not substitute an illustrative or cached sample.
Current all-time token totals for listed public participants; identities are not sent to the browser calculator.
Explains what the CLI aggregates, what it does not collect, and the limits of the public cohort.