Monitor this situation.
Unsubscribe anytime.
[SITUATION] · [QUIET] · [TECHNOLOGY]
2 clusters · 6 sources · 27 days · First seen · Last updated
Vercel AI Gateway model usage trends
Overview
Data from the Vercel AI Gateway indicates a significant shift in AI model usage, characterized by a growing divide between token volume and financial expenditure.
In late August 2026, open-weight models accounted for 62% of all tokens processed, a sharp increase from just 11% in April. By mid-September, this volume rose further to approximately 78.4%. This surge is largely driven by Chinese models, specifically DeepSeek, which has emerged as a leader in volume. For instance, the DeepSeek V4.1 Flash model was reported to account for 59.3% of processed tokens.
Despite the dominance of open-weight models in terms of volume, proprietary models from companies like Anthropic continue to capture a majority of total expenditure. This has created a “barbell pattern” in enterprise usage: massive volumes of low-cost inference via open-weight models contrasted with high-cost, premium workloads using closed-source models. This economic trend is supported by a rapid decline in token prices, which fell by over 50% in the months leading up to August 2026.
Entities
Timeline
-
[TECHNOLOGY] 2 sourcesVercel AI Gateway data shows split between model volume and spend
Vercel AI Gateway data shows open-weight models like DeepSeek dominating token volume, while closed models like Anthropic’s Claude command the majority of total AI spend.
-
[TECHNOLOGY] 5 sourcesVercel reports open-weight AI models hit 62% of gateway traffic
Open-weight AI models reached 62% of token volume on Vercel’s AI Gateway in August, driven by cost-reduction strategies and the popularity of models like DeepSeek.
Sources
chinatalk.media · cryptobriefing.com · gizmodo.com · ictbusiness.biz · ontheissuesmagazine.com · scmp.com