Priciest tokens still pull the most gateway spend

Anthropic took 65.1% of the money spent through Vercel’s AI Gateway in July while carrying 30% of its tokens, TechRadar reported on August 17 from Vercel’s monthly data. The two shares measure different things. Anthropic handles less than a third of the gateway’s traffic, but its tokens cost enough that it still collects most of the gateway’s spend.
The gateway is an API service that routes calls between model providers, so customers can swap one lab for another without much work. Vercel has reported the spend split every month since December 2025, and Anthropic has held more than 60% of it in each of those months. In July its average price per token was 4.4 times higher than the overall average across all AI labs. The report defines price per token as total spend divided by total volume.
Prices across the gateway moved the other way from spend in July. Total spend rose 37%, while the average price per token fell 13.6%, after a 20% rise in May and little change in June. Anthropic’s own per-token average dropped as well, a change attributed to Claude Fable 5 coming back to public access after its temporary suspension.
That suspension followed a US export-control directive on June 12 that barred any foreign national, inside or outside the United States, from using Fable 5 and Claude Mythos 5. Reliably filtering people by nationality in real time was not practical, and a narrow block would have cut off many legitimate users, including Anthropic’s own foreign-born staff, so the company pulled both models for every customer worldwide. The Commerce Department removed the controls on June 30, and Anthropic redeployed both models globally on July 1 with an improved safety classifier aimed at the jailbreak technique the directive cited.
The “Apple of AI” label comes from TechRadar. The outlet reads Anthropic’s larger revenue pull at a higher token price as a sign that recent versions of Claude, along with anticipation over the gradual release of Mythos 5, have given the company a strong reputation.
Further down the July table, DeepSeek, which had been growing since May, passed Google to take second place by volume. Vercel’s report had predicted it: “we said an open-weight lab would soon be second by volume. In July, DeepSeek surpassed Google to take that place.”
Up next
Hidden China tracker pulled from Claude Code after exposé
“Hiding the signal in the system prompt makes every other privacy claim harder to believe,” wrote the developer who exposed Claude Code’s tracker for Chinese users. Anthropic says it was a March experiment, now removed.
Related stories
- Investor update puts run rate at $65B before IPO
- GPT-5.6 Luna output drops from $6 to $1.20 per million
- Claude’s free Team year for startups lasted only days
- Claude Startups widens its door, with up to $45K in perks
- Claude plans can eat over 40% of Anthropic’s compute
- Cheaper rivals outdraw Anthropic’s best model, FT says
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

