AI models are becoming outdated at record speed — so what does that mean for the tokens that companies are spending millions on?
Vercel claims models older than three months are being ditched in favour of newer iterations, with older versions considered outdated and effectively retired
Artificial intelligence models are updating at an alarming pace, rendering older versions obsolete within months. Vercel’s AI Gateway reports that most AI tokens are spent on models that have only been available for around three months, and as usage increases, token prices are falling. This trend is hastened by the popularity of open-weight models, which allow users to self-host or run on affordable cloud servers, making them a more cost-effective alternative to closed, black-box systems.
Open-weight models are gaining traction, with Vercel’s report indicating that they now make up 56% of token volume, up from just 13% in April 2026. This is a significant shift from December 2025, when open-weight models represented only 10% of token use. Among the open-weight models gaining attention are Meta Llama 4 and Jev from TypeSafe AI, which has made a notable impact with its low token cost and rapid adoption.
However, the rapid development and release of new AI models may lead to a change in token spend. As companies commit to newer models, the investment in older, legacy models – where tokens are locked to specific versions – becomes a contentious issue. This debate highlights the rapid pace of AI development and its potential implications for token pricing and usage.
Written by urgent.news from TechRadar's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
