open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
Why it matters
It gives you empirical production numbers — not vendor benchmarks — on which models are actually absorbing tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → volume versus where the money goes, useful when deciding whether to route cheap open weightsA model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.Full definition → models for bulk work and reserve premium models for the paths that need them. The per-modality breakdown also shows what real teams are picking for image and video generation.
Key quotes
“Open-weight models ran 29% of gateway tokens in June on just under 4% of spend. That is nearly a third of the tokens for one twenty-fifth of the dollars.”
“This is evidence of the routing discipline June's report documented, now visible in the aggregate: high-volume work goes to low-cost models, high-risk work stays on the frontier.”
“Roughly one in eight enterprise customers now runs an open-weight model in production.”
“And on current trajectories, an open-weight lab will soon be the second-largest by volume on AI Gateway.”
“The market spent more overall, but not more per token.”
