Start here

AI coding leaderboard for tokens and streaks

See how public UsageAtlas profiles rank by AI coding tokens, streaks, and consistency. Filter by Claude Code, Codex, Cursor, model, and period.

Quick answer

A leaderboard ranks people or tools by a metric. Ranking by raw tokens is the most common version and the least useful, because long context and retries inflate it. If you want a leaderboard, rank completed tasks per unit of spend, or rank tools by cost per finished task.

Pick a metric that means something

MetricRewardsFailure mode
Total tokensLong sessionsLoops and retries
Tasks completedShippingTrivial tasks
Cost per taskEfficiencyAvoiding hard work
Blocked days avoidedPlanningLow ambition
Model diversityExperimentationNovelty

If you run one internally

  1. Make participation opt-in and aggregate-only.
  2. Rank teams or tools, not individuals, in the first version.
  3. Publish the formula next to the ranking.
  4. Reset monthly so a bad week is not permanent.
  5. Kill it if behaviour gets worse.

The point of a leaderboard is learning, not scoring. A ranking that makes people avoid hard tasks has negative value no matter how accurate it is.

Rank the tools instead

Ranking tools by cost per completed task is more useful and less political. UsageAtlas gives the token side for 14 providers, and cost per project gives the formula.

For a shareable summary rather than a ranking, see AI coding wrapped.

Design a board that rewards finished work

Token leaderboards reward long sessions, which is the opposite of what most teams want. If you publish one, pair it with a second column: tasks completed, review comments resolved, or tests made green. The pairing removes the incentive to leave an agent running for the sake of a higher number.

Opt-in matters too. A public ranking of individual token use changes behaviour, and not always for the better. Team-level aggregates answer the budget question without turning a cost report into a scoreboard. The wrapped guide covers the summary format.

Questions, answered

What is an AI coding leaderboard?

A ranking of people or tools by a usage metric. Token totals are common but reward long sessions and retries rather than useful output.

Are leaderboards a good idea for teams?

Only if they rank teams or tools on cost per completed task. Per-person token ranks distort behaviour.

Is there a public leaderboard?

Several community boards exist, but their totals are self-reported and not comparable across tools. Treat them as entertainment.

Browse by topic