Intel / article

Open-weight models take 56% of token volume, Astra doubles Fable 5.1 spend

Source: vercel.com
Author: Amelia Charles
Date: Published Sep 17, 2026

Context

Model usage and model spending tell different stories in Vercel's September AI Gateway index. models, whose trained parameters are published for anyone to download, processed 56% of August's volume, while Anthropic received 64% of estimated spending. The report counts input, output, reasoning and cached text units in volume, and estimates spending from labs' published list prices, so actual bills may differ. A model category can therefore handle most of the text without capturing most of the money. These figures describe anonymized traffic through this one gateway: they show open-weight models carrying the majority of volume while spending stayed concentrated with Anthropic, rather than a single winner on every measure of adoption.

Key takeaways · AI-distilled

Open-weight models ran the majority of AI Gateway token volume for the first time in August 2026 — 56%, up from just 7% in December 2025 — even as frontier labs kept most of the dollar spend.
Average price per token on the gateway fell 23.2% in August, the third straight monthly drop, and teams running over 10 million tokens paid a median 7.6% less per token than the prior month, more than double July's 2.9% decline.
Nine in ten teams using Claude Fable 5 cut their usage after Opus 5 (half of Fable's price) launched, and most of that spend moved to Opus 5 rather than a rival lab — so Anthropic kept roughly 61-64% of all gateway spend every month since December despite the churn.
GPT-6 Astra took a third of OpenAI's own gateway spend within two days of its September 3 launch, and over each model's first twelve days it out-earned Anthropic's Fable 5.1 more than 2x in gateway spend share (7.7% vs 3.7%).
Google's Gemini 3 Flash lost 95% of its gateway token share since May, split roughly evenly between cheaper models (led by GPT-5.6 Luna) and pricier ones (led by Claude Opus 5 and Sonnet 5) — pushing Google's total gateway token share from 30% to 5%.

Terms in this piece · Glossary

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.

Read the source vercel.com