Open-weight models win hearts, minds, and wallets
Open-weight AI models gained ground on Vercel’s AI Gateway from June to September, handling 71.5 per cent of tokens and nearly a quarter of spending. The shift matters because it points to growing commercial use of these models, while possible US restrictions and new Chinese export controls could affect businesses that host or rely on them.
Open-weight models’ share of gateway spending rose more than fourfold, from about 5 per cent to nearly 25 per cent; gateway spending overall was about 13 times December 2025’s level, while average token prices fell 47 per cent. Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6 Astra led spending, while DeepSeek V4.1 Flash ranked third. Almost one in three teams changed its most-used provider in September, and Vercel said more than two-thirds of spending on Chinese models went to US providers serving them.
- Open-weight models handled 71.5 per cent of Vercel gateway tokens.
- Their share of spending rose more than fourfold from June to September.
- Possible trade controls could affect US businesses serving Chinese models.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
The figures suggest open-weight models are becoming commercially useful at scale, handling most tokens while taking a growing share of spending. Their availability can give businesses more choice over hosting, cost and customisation, and reliance on US providers to serve Chinese models shows that firms are already finding ways to use them within existing services.
The case against
Gateway usage and spending do not by themselves show that open-weight models are more reliable, secure or suitable for every business task. Their growth also raises concerns about dependence on models and supply chains exposed to changing US restrictions or Chinese export controls, while frequent provider switching and falling token prices may make current market trends hard to interpret.