OpenRouter's 2025 usage study found that reasoning-optimized models accounted for more than half of the text processed through its platform, up from a negligible share early that year. This measures traffic routed to those models, including input and output text; it does not isolate their internal reasoning. The report therefore documents a shift in model selection within one service. Its announcement describes analysis of metadata covering more than 100 trillion tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition →. Locally hosted models and other traffic outside the platform remain beyond the study's scope.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
