The shift is dramatic by pace: in December 2025, open-weight models handled fewer than one in ten tokens. By April 2026, their share was 13%. By August, it exceeded 56%. Their share of spending tells a different story — open-weight models accounted for 14% of gateway spend in August, against 86% for closed-weight models.
Token prices have followed. The average cost per token fell 23.2% in August, the steepest single-month decline since April 2026 and the third consecutive monthly drop. The median team running more than 10 million tokens per month paid 7.6% less per token in August, against a 2.9% decline the previous month.
Anthropic's Fable 5 lost roughly two-thirds of its gateway spend share as teams moved to the cheaper Opus 5, but Anthropic as a lab retained 64% of gateway spend because those teams stayed within its model lineup. Anthropic has held at least 61 cents of every gateway dollar each month since December 2025, with its models occupying the top two spend positions throughout.
OpenAI launched GPT-6 Astra on 3 September at the same price point as Fable 5.1 and roughly 2.5 times the price of GPT-5.6 Sol. Within two days, Astra took one in three dollars of OpenAI's gateway spend — a faster share uptake than Fable 5.1 achieved in its own first 12 days. Across the period from 4 to 16 September, Astra and GPT-5.6 Sol together handled 27% of OpenAI tokens but generated 71% of its spending.
Half of all gateway tokens in August ran on models that had been available for fewer than three months. Nearly nine in ten tokens used models that arrived on the gateway within the previous four months.
Vercel AI Gateway routes tens of trillions of tokens monthly between production applications and AI labs, providing one of the cleaner views of actual enterprise consumption patterns.
To stay across the latest in cloud, AI and enterprise tech analysis from Compare the Cloud, subscribe to our weekly newsletter at https://www.comparethecloud.net/newsletter
