Vercel Data Shows AI Model Churn Is Accelerating as Open-Weight Models Gain Ground
Vercel’s AI Gateway data show most AI tokens are spent on models less than four months old, open-weight models account for 56% of token volume, and average token prices fell 23.2% for a third straight month.
The AI Gateway Production Index found that the average price per token fell for the third consecutive month, dropping 23.2% in the latest period. TechRadar reported that the decline stems from multiple factors, but a key influence is the spread of token pricing across the industry. Open-weight models can be self-hosted or run on affordable cloud servers, giving companies alternatives to closed models.
Vercel’s September index shows open-weight models took 56% of token volume but only 14% of token spend. That marks a rapid shift: open-weight models had just 13% of token volume in April 2026 and 10% of token use in December 2025. The report says enterprise AI users are increasingly recognizing that a closed model can be swapped for a more affordable and capable open-weight alternative.
The competitive pressure is visible among named models. Fable 5, described as Anthropic’s most capable model, lost out to Anthropic’s own Opus 5 and OpenAI’s Astra, according to the Vercel data. Newer entrants are also having an effect, including TypeSafe AI’s Jev, which Vercel highlights as leaving a noticeable footprint in token pricing.
Jev has a very low token cost but relies on a different type of input: data rather than conversation. Vercel also notes that Jev is the fastest-adopted model since the AI Gateway Production Index was launched. Meta Llama 4 is among the open-weight models making an impact, according to the report.
The churn in model versions is changing how spending works. Although improvements in AI models can produce better results for productivity and other uses, the faster development and release of newer models creates uncertainty for buyers who purchase tokens tied to older or legacy models. TechRadar reported that this commitment to older models highlights one of the biggest debates in AI: that development is moving too quickly.
Vercel’s findings suggest the market is rewarding newer models and open-weight options at the same time. Token spending is increasingly directed at models that have only been available for the past three months, while the average token price continues to fall. The shift has implications for companies that have committed budgets to models that may be replaced within months, and for model providers competing on both capability and cost.