Open-weight models take 56% of token volume, Astra doubles Fable 5.1 spend
Open-weight models just hit 56% of production token volume. Astra took a third of OpenAI's spend in 48 hours. Here's what actually moved in August.
Why it matters
Real production traffic data reveals the frontier lab race is reshaping around price-to-capability fit, not brand loyalty. Open-weight surge, Gemini collapse, and Astra's immediate dominance over Fable show which models teams actually choose when it matters.
The key facts
9 to knowOpen-weight models: 7% token volume in December 2025 → 56% in August 2026 (first month at majority)
Frontier models still hold majority of spend despite volume shift; Anthropic retained 64 cents of every dollar spent on AI Gateway in August
GPT-6 Astra: 7.7% of all gateway spend in first 12 days vs Fable 5.1's 3.7%; 2x team adoption
Fable 5: 13.2% spend share (July) → 4.9% (August) after Opus 5 launch; 90% of teams running Fable cut usage
Opus 5: gained 22.5% spend share in August; tripled its share despite half Fable's price
Google Gemini 3 Flash: 30% token volume (May) → 5% (August); lost 95% of share since May
Average price per token: fell 23.2% in August (third consecutive monthly drop); median team paid 7.6% less per token
Token price now less than half what it was five months prior
Data source: Vercel AI Gateway (anonymized, aggregate production traffic through August 2026)
Go to the source
Vercel Blogvercel.com
Publisher excerpt: AI Gateway Production Index — September 2026 Every month, routes tens of trillions of tokens between production applications and AI labs. That traffic gives us a view of what AI usage actually looks like in today's enterprise, and we publish it here monthly. See the Production Index reports from ,…