Open weight models process majority of AI tokens

Open weight AI models have processed the majority of AI tokens for the first time, according to Vercel’s AI Gateway Production Index. In August, open weight models accounted for 56% of all tokens processed via the gateway, a significant increase from 13% recorded in April.

Despite the high volume of usage, open weight models represented only 14% of total spending through the AI Gateway. This trend coincides with a 23.2% decline in the average price per token in August, marking the third consecutive monthly decrease.

Usage patterns within Anthropic’s offerings also showed a shift toward more cost-effective options. While Claude Fable 5’s share of total spending dropped from 13.2% in July to 4.9% in August, Opus 5’s share grew to 22.5%, as the latter costs approximately half as much per token as Fable 5.