Vercel reports open-weight AI models hit 62% of gateway traffic

Open-weight AI models have seen a massive surge in usage on Vercel’s AI Gateway, accounting for 62% of all tokens processed in late August. This represents a significant increase from April, when open-weight models held only 11% of the platform's token volume.

Data indicates that developers are increasingly opting for lower-cost models for production workloads, such as autonomous agents, to reduce expenses. While open-weight models dominate in token volume, proprietary models from companies like Anthropic still capture a majority of the total expenditure on the gateway due to their higher costs.

Chinese models, led by DeepSeek, have risen to the top of the token volume leaderboard, overtaking Google in processing share. DeepSeek-V4-Flash was identified as the most-used model by volume, with other Chinese models like Step 3.7 Flash and GLM-5.2 also appearing in the top five rankings.