Benchmarks tell you which model is smartest; this tells you which one developers actually pay to run in production. The gap is often huge — a model can top every intelligence leaderboard yet sit seventh by real usage, while a cheaper rival quietly captures the bulk of the traffic. Because it draws on aggregated token flow across millions of users and 400+ models behind one API, it functions as a live market signal for the LLM industry rather than another scoreboard.
What Sets It Apart
- Ranks by revealed preference — weekly token volume of paying traffic — so price/performance and reliability are baked into the numbers, not just lab-reported scores.
- Slices the same data several ways: top models, token market share by lab, tool-call frequency, and image-processing volume, letting you see how models are used, not just that they're used.
- Captures macro shifts early: open-vs-closed splits, the rise of Chinese-origin labs, and quarter-over-quarter throughput growth show up here before they hit headlines.
- Vendor-neutral by construction — OpenRouter routes across 60+ providers, so no single lab's marketing skews the ranking.
Who It's For
Great fit if you're choosing a model for a real workload, sizing a market, or tracking adoption trends and want evidence over vendor claims. Look elsewhere if you need controlled capability evaluations — this measures what people deploy, which is shaped by cost, latency, and availability as much as raw intelligence. It also only reflects traffic that flows through OpenRouter, so models served mainly via first-party APIs are underrepresented.
