"Best available public data source" isn't a good reason to insist on using irrelevant data to draw conclusions like "top 5 models by token consumption are Chinese/open models".