Vercel AI Gateway data shows split between model volume and spend

Data from the Vercel AI Gateway reveals a diverging trend between AI model usage volume and expenditure. Open-weight models have seen a significant surge, accounting for up to 78.4% of token volume in recent leaderboard data, a substantial increase from 11% in April 2026.

DeepSeek has emerged as a leader in volume, with its V4.1 Flash model accounting for 59.3% of tokens processed through the gateway. In contrast, closed-source models like Anthropic’s Claude Opus 4.8 dominate in terms of financial spend, capturing 13.7% of total expenditure despite representing a much smaller fraction of token volume.

This creates a barbell pattern in enterprise AI usage: massive volumes of low-cost inference driven by open-weight models, and premium, high-cost workloads utilizing closed models for specific quality requirements. The economic shift is further driven by a rapid decline in token prices, which fell over 50% in the five months leading up to August 2026.