Vercel’s AI Gateway numbers paint a striking picture of who’s actually making money in the model wars, and the answer is unambiguous: Anthropic. The company pulls in 65.1 percent of all Gateway spending while accounting for just 30 percent of tokens served. Do the math and you land on a per-token cost that’s 4.4 times the average of every other provider on the platform. Developers aren’t just using Claude — they’re paying a premium for it, and they keep coming back.
That willingness to pay extends to newer entrants, too. Fable 5, a model that barely registered a few months ago, has shot up to 13.2 percent of Gateway spending, second only to Opus 4.8. What’s more telling: nine out of ten Fable teams in July were brand-new customers. Vercel reads that as evidence the model is pulling fresh users into the ecosystem rather than cannibalizing existing spend. It’s a notable counterpoint to payments company Ramp, which had previously flagged low Fable usage as a signal the model might be priced too aggressively. Both companies are looking at different slivers of a much larger market, so neither has the full picture.
Overall token volume on Vercel climbed 59 percent, while spending grew 37 percent — a gap explained by average token prices falling 13.6 percent. Companies are clearly shifting toward cheaper tiers and smaller models where they can. The pattern mirrors what we’ve seen across the industry since late 2023: raw usage keeps climbing, but unit economics keep compressing for everyone except the models developers genuinely can’t replace.
The Anthropic premium is the real story here. Claude’s per-token pricing on Vercel isn’t just high relative to budget models — it’s high relative to everything. That developers keep paying suggests Anthropic has built something closer to a moat than most observers would’ve credited a year ago. Whether that moat survives as open-weight models keep improving is the question nobody can answer yet, but the revenue numbers say the market isn’t waiting around to find out.
