Over the 30 days ending July 7, 2026, the public models ranked on OpenRouter processed about 197.18 trillion tokens of inference — roughly 6.57 trillion tokens per day. This is a demand-side context metric for the agent economy (off-chain), not on-chain activity: agent transactions are downstream of the inference that powers them. Source: OpenRouter (openrouter.ai/rankings).
Daily AI inference demand (trailing window) · T tokens/day
Daily AI inference demand (trailing window), in T tokens/day
Period
T tokens/day
06-07
5.31
06-08
6.53
06-09
6.43
06-10
6.7
06-11
6.93
06-12
6.77
06-13
5.67
06-14
5.58
06-15
6.4
06-16
7.02
06-17
7.76
06-18
7.31
06-19
6.56
06-20
5.83
06-21
5.78
06-22
6.92
06-23
7.15
06-24
7.25
06-25
6.97
06-26
6.78
06-27
5.69
06-28
5.95
06-29
7.09
06-30
7.17
07-01
6.98
07-02
6.97
07-03
6.99
07-04
5.71
07-05
5.81
07-06
7.15
Demand-side context, not on-chain data
Every other number on this site is measured on-chain. This one is not: it is a demand-side context metric — how much large-language-model inference the public models ranked on OpenRouter are serving — included because agent activity is downstream of inference. When token throughput grows, the population of running agents that could eventually transact on-chain grows with it. It is context for the on-chain series, not a substitute for them.
The figure sums daily token usage across the top-50 public models on OpenRouter plus an aggregated “other” row, over a trailing default window. It captures only traffic routed through OpenRouter — a large, public, but partial slice of all inference — so read it as a directional demand indicator, not a census of every AI token served.
Why the token totals are not comparable across models
Different model families use different tokenizers, so one provider's token is not the same unit of text as another's. Summing tokens across models therefore mixes units — the total is a useful trend line for aggregate demand, but it is not a precise, apples-to-apples count, and per-model comparisons drawn from it would mislead. The daily series below is most valuable read as a shape over time, not as an exact quantity.
Also asked
Is this the total amount of AI inference in the world?
No. It counts only inference routed through OpenRouter across the public models it ranks — a large but partial slice. It also mixes model-specific tokenizers, so the total is a directional demand indicator, not an exact, universal token count.