Open-Weight AI Models Capture 56% of Volume as Token Prices Fall 23%

Data from Vercel’s AI Gateway index reveals rapid model turnover and a surge in open-weight adoption that is upending enterprise AI spending.

By The Company WireSeptember 28, 2026 at 6:05 PM UTC3 min read

Vercel — Open-Weight AI Models Capture 56% of Volume as Token Prices Fall 23%. Photo: TechRadar Pro.

Enterprise spending on artificial intelligence is shifting toward newer architectures at record speed, according to September data from Vercel’s AI Gateway Production Index reported by TechRadar Pro (https://www.techradar.com/pro/ai-models-are-becoming-outdated-at-record-speed-so-what-does-that-mean-for-the-tokens-that-companies-are-spending-millions-on). The report indicates that the operational lifespan of deployed AI models is shortening rapidly, with token usage concentrated heavily in models released within the past three months.

The index recorded a 23.2% decline in the average price per token in September, marking the third consecutive monthly price drop. Vercel attributed the deflationary trend to broader industry price competition and the expanding use of open-weight models, which allow companies to host workloads independently or deploy them on lower-cost cloud compute.

Open-weight models accounted for 56% of total token volume processed through Vercel's gateway in September, but represented only 14% of overall token spend. That marks a sharp rise in utilization over recent months: open-weight architectures accounted for 10% of token volume in December 2025 and 13% in April 2026 before climbing past half of total volume.

The shift has triggered notable churn among proprietary and open models alike. Vercel’s index showed Anthropic’s Fable 5 losing ground to newer models including Anthropic’s Opus 5 and OpenAI’s Astra. Meanwhile, Meta’s Llama 4 and upstart options such as TypeSafe AI’s Jev are driving down blended infrastructure costs. Jev, which processes structured data inputs rather than conversational prompts, was identified as the fastest-adopted model since the AI Gateway Production Index launched.

The compression of model release cycles has also complicated enterprise procurement. Long-term token commitments or upfront volume purchases tied to specific legacy models present mounting operational risks as cheaper, higher-performing models regularly displace incumbents within weeks of release.