Insights / Attention share of OpenRouter traffic
Insights · Market

Who's getting the tokens?

Share of OpenRouter token volume, daily, classified against the LLMap catalog — by weights class and by lab HQ country.

Open models
41%
US labs
23%
China labs
45%
Open vs closedUS labs vs China labs vs rest

Latest day in the series

On Wed, Sep 30, open-weight models took 41% of the tokens routed through OpenRouter; closed models took 49%, and 10% went to models we have not matched to our catalogue.

Open-weight modelsClosed modelsNot matched to a model we track
0%25%50%75%100%Jun 28Jul 12Jul 26Aug 9Aug 23Sep 6Sep 20Open-weight models41%Closed models49%Not matched to a model we track10%

Hatched: tokens we could not match to a model we track, so they are neither open nor closed to us. Point at any day to rewrite the sentence for it.

Source: OpenRouter (openrouter.ai/rankings), as of 2026-09-30 · weights & country classification: LLMap catalog · token counts use each provider's own tokenizer — read as relative attention share, not absolute compute
Cite thisLLMap (2026). "Attention share of OpenRouter traffic: open weights vs closed." llmap.ai/insights/market-share. Data as of 2026-09-30. Source: OpenRouter (openrouter.ai/rankings), as of 2026-09-30.
CSV
CC BY 4.0 — reuse freely with the citation aboveupdated daily · data to Sep 30