Open weight takes 71.5% of tokens and 23% of spend

This post details the increasing adoption and spend share of open-weight AI models through Vercel's AI Gateway, highlighting their rapid growth and cost-effectiveness compared to frontier models. It quantifies the shift in market dynamics, showing open-weight models now processing over 70% of tokens and accounting for nearly 25% of spend. Specific models like DeepSeek-v4.1-Flash and Kimi K3 are analyzed for their token volume and cost impact. The post also reveals a trend of US-based providers serving a significant portion of spend for Chinese-origin models, indicating a complex global AI infrastructure. It notes record levels of model and lab switching by teams, suggesting a highly competitive and fluid AI model market.