How to read this page
What does this AI usage data measure?
Each figure counts the tokens, input plus output, that AI models processed through OpenRouter, a service developers use to reach models from many labs through one API. A token is a small chunk of text, often described as about three quarters of an English word. Counts are totalled per UTC day, and each release or variant of a model is counted separately.
Does this show how much the industry spends on AI?
No. It shows tokens, not dollars, and it only covers public traffic routed through OpenRouter. Direct contracts with a lab, subscriptions such as ChatGPT or Claude, and cloud platforms such as Amazon Bedrock or Google Vertex AI are not included, and OpenRouter excludes private and zero-data-retention traffic before publishing. Read it as a signal of which models developers choose and how fast that demand is growing.
Why can a lab have a large token share but a smaller share of revenue?
Prices per token can differ by more than tenfold between models, so a lab with low-priced models can process far more tokens than a rival while earning less from them. Token counts also come from each provider's own tokenizer, so a token on one model is not exactly the same as a token on another.
How are lab totals calculated?
OpenRouter publishes the 50 most-used models each day plus one row for everything else. Lab totals here are sums of those daily top 50 rows, so a lab's smaller models can fall into the row labelled outside top 50, and changes for models near the cutoff are approximate. OpenRouter can restate recent figures, so recent days are refreshed at every update.
Source: OpenRouter (openrouter.ai/rankings), as of Nov 5, 2026. Licensed under CC BY 4.0. Lab totals, shares and changes are calculated by Frenzy Capital from that data. This page is for information only and is not financial advice. All trademarks belong to their respective owners.