Open-weight AI models capture 71.5% of Vercel Gateway token volume

Open-weight artificial intelligence models processed 71.5% of all tokens through Vercel’s AI Gateway in September, according to data released by the company. This marks a significant shift in market dynamics, as these models now handle the majority of token volume on the platform for the first time.

The financial impact is also growing rapidly. According to Vercel, the share of spending allocated to open-weight models has more than quadrupled over the last three months. In June, open-weight models accounted for one in twenty gateway dollars, but by September, that figure rose to nearly one in four. This indicates that while proprietary models still command higher prices, the volume of usage for open-weight alternatives is expanding significantly.

Market Dynamics and Pricing Trends

Vercel’s AI Gateway functions as an intermediary, allowing customers to purchase tokens from various AI providers through a single interface. This setup supports both proprietary and open-weight models, though it does not track usage from direct vendor connections or on-premises hosting. The data reveals a broader trend of cost efficiency and flexibility among users.

Total spending on the gateway in September was approximately 13 times higher than in December 2025. Despite this surge in expenditure, the average price per token dropped by about 47 percent, according to Vercel. This price reduction, combined with the rise in open-weight usage, suggests that market competition is intensifying. Nearly one in three teams changed their most-used AI provider in September, indicating that vendors are facing pressure to retain customers.

Model Popularity and Regional Shifts

TypeSafe AI’s Jev model emerged as the most popular choice among teams in September. In just 14 days, it reached 35 percent of paid teams using the Vercel AI Gateway. However, its popularity did not translate into significant spending. Due to its architecture, Jev accounted for only 1 percent of the month’s tokens and a mere 0.15 percent of total spending.

US-based models from Anthropic and OpenAI continue to attract the largest portion of financial investment. Claude Opus 5.5 led in spending with 11.03 percent of the total, followed by GPT-6 Astra at 4.86 percent. Meanwhile, models from Chinese providers are gaining traction. DeepSeek V4.1 Flash ranked third in spending with 4.19 percent, highlighting a diversification in the sources of AI technology being utilized by enterprises.

What should you do?

Security and technology professionals should monitor their organization’s AI spending and usage patterns closely. As the market shifts toward open-weight models, it is essential to understand the security implications of these alternatives. Teams should evaluate whether their current providers offer sufficient transparency and control.

What to do and how to stay safe: Vercel

• Review your current AI provider contracts and assess potential vendor lock-in risks.
• Evaluate the security posture of open-weight models before integrating them into production environments.
• Monitor token usage and spending trends to identify cost-saving opportunities without compromising security.
• Stay informed about new model releases and updates from both proprietary and open-weight providers.

Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What percentage of Vercel AI Gateway tokens did open-weight models handle in September?

Open-weight models processed 71.5% of all tokens through Vercel’s AI Gateway in September.

How has spending on open-weight models changed recently?

Spending share for open-weight models more than quadrupled from June to September, rising from one in twenty dollars to nearly one in four.

Which model was the most popular among teams in September?

TypeSafe AI’s Jev model was the most popular, reaching 35% of paid teams using the gateway in just 14 days.