Vercel's AI Gateway index shows open-weight models took 56% of token volume but only 14% of spend
Vercel's September AI Gateway Production Index, published 17 September on August data, reports open-weight models crossed a majority of token volume for the first time at 56%, up from 7% in December 2025, while accounting for just 14% of dollars spent. Anthropic still took 64% of all gateway spending, with Claude Opus 5 alone at 22.5%. Average token cost fell 23.2% in August and more than 50% over five months, and in matched first-twelve-day windows GPT-6 Astra took 7.7% of gateway spend against Fable 5.1's 3.7%, used by twice as many teams. This is the clearest public split yet between where tokens go and where money goes, and it says open weights are winning the cheap high-volume work while frontier closed models keep the expensive calls.
Source
↳ Follow the thread