Token share measures how much of a defined pool of AI text processing went through a model or provider. If a platform records the input and output tokens processed by each model, it can divide one model's total by the total across the measured models for the same period.
It differs from market share based on revenue, number of customers or count of requests. A few customers running very long prompts can account for many tokens, while a popular model used for short answers may have a smaller token share.
A dataset may cover only one API router, company, region or collection of participating applications. Some estimates count generated output separately from input or exclude cached tokens. As a result, two published token-share figures may describe different markets and should not be combined casually.
Before treating a token-share chart as evidence of open-versus-closed adoption, check the denominator, time window, data source, token categories and whether different models are grouped consistently.