What the data says
25th percentile
1.07B
One quarter of observed totals are at or below this value.
Median
4.30B
Half of observed totals are at or below this value.
75th percentile
10.36B
One quarter of observed totals are above this value.
Top-decile threshold
28.24B+
The smallest total among the highest 10% of rows, rounded up.
Up to 1.07B
At or below the 25th percentile
1.07B–4.30B
Above Q1 through the median
4.30B–10.36B
Above the median through Q3
10.36B–28.24B
Above Q3 and below the top-decile threshold
28.24B+
At or above the top-decile threshold
Token totals measure volume, not code quality, productivity, or value delivered. A large repository, a long submitted history, a model with heavy prompt caching, or an agent that records more complete logs can all raise a developer's total. Compare the result with your own time window and workflow before treating a percentile as meaningful.
The most useful comparison is directional: determine whether your total is below the middle, in the interquartile range, or in the upper tail, then inspect which agents and models created it. The benchmark is not an allowance, billing limit, or recommendation to burn more tokens.
Separate 30-day context
This is a cohort-wide 30-day activity total, not the distribution used above. The percentile benchmarks use each listed developer's all-time submitted total so one number is available per participant. Keeping the scopes separate avoids presenting a lifetime leaderboard as a monthly usage benchmark.
Population. The calculation includes current public leaderboard rows with a finite, positive all-time token total. Private, unlisted, suppressed, and local-only users are not represented. One row contributes one observation; developers are not weighted by activity.
Percentiles. Totals are sorted from smallest to largest. Quartiles use the nearest-rank observation at ceil(percentile × sample size). The top-decile threshold is the smallest value among the highest ceil(10% × sample size) rows. No interpolation is used, so every reported threshold is an observed total. Ties can make the top group larger than exactly 10%.
Coverage. The sample includes only people who chose to appear publicly and whose supported local usage was submitted. It is an opt-in product cohort, not a representative survey of software developers or all AI coding-agent users. Cross-tool logging completeness can differ.
Freshness. The displayed cohort snapshot was generated Aug 28, 2026, 9:16 AM UTC. Values can move whenever participants sync, change visibility, or are added to or removed from the public board.
Current listed-developer rows used to calculate the all-time distribution and thresholds on this page.
Aggregated model, agent, trend, skill, and tool-call totals from participating developers.
Explains what the CLI aggregates, what it does not collect, and the limits of the public cohort.