The monthly AI Gateway index for August: open models handle most tokens for the first time, the price per token falls for the third month in a row, and Anthropic takes 64 cents of every dollar.
- Open models: from 7 per cent in December to 56 per cent of tokens in August.
- Fable 5 loses two-thirds of its spend share in a month, while Opus 5 at half the price triples it.
- The data are Vercel's for its own gateway, not for the whole market.
Most of the work, less of the money. That is the August snapshot of a gateway that routes tens of trillions of tokens a month between applications and models.
56 per cent of tokens go through open models, and 14 per cent of spend goes to them. The gap between the two numbers is the whole story.
What does not match July
In its July report Vercel gave 11 per cent open models for April. In the current one, the same April is 13. The difference is small, and Vercel itself explains it: it now counts a broader set of models as open and revises earlier months. It is worth knowing, when you compare the two texts, that the numbers move backwards too.
The really interesting part is Fable against Opus. Nine out of ten teams that used Anthropic's most expensive model cut their usage. Most often they went not to a competitor, but one step down in the same shop.
For anyone paying a model bill, the takeaways are two and both are practical. Put the cheap open model where the work is high-volume and simple. Try the step down on the expensive one before deciding you need the top. Nine in ten teams on this gateway that ran Fable have already cut back on it.