Question 1

Where does the usage data come from?

Accepted Answer

OpenRouter's public rankings page (openrouter.ai/rankings), which OpenRouter publishes under CC BY 4.0. OpenRouter is a model-agnostic aggregator routing over 100 trillion tokens a week across every major frontier and open-weight model, so its rankings are the closest public proxy for "what model is the market actually picking." Each dated snapshot takes the top 20 of OpenRouter's rolling 7-day window of complete UTC days and republishes it in machine-readable form with OpenRouter's citation line. The snapshot date and window are shown on this page and in the API response.

Question 2

How is this different from /benchmarks or /attention?

Accepted Answer

Benchmarks measure synthetic capability. Attention measures news mentions. Usage rankings measure real production traffic. The three are complementary and often disagree: a model can win benchmarks but lose usage if it is too expensive, or trail benchmarks but win usage if it is cheap and reliable.

Question 3

Which model leads the usage rankings?

Accepted Answer

The leader can change between snapshots, so check the table on this page or /api/usage-rankings for the order in the current snapshot window. Token volume tends to reward cheap, capable models that agents can call in high-volume loops, which is why a model can trail the frontier on benchmarks and still top the usage table.

AI Usage Rankings