Skip to content

An editorial team and a community building with AI

← All articles

News · Vercel · 2 min read

Event date: · Published:

Vercel AI Gateway report: decision models hit 26% of requests

Vercel's November 2026 AI Gateway Production Index shows decision models reaching about a quarter of requests and Anthropic winning new teams nearly two to one over OpenAI in October 2026.

Vercel published its AI Gateway Production Index, November 2026 report on November 9, 2026, covering AI Gateway usage data from October 2026. The report found that lightweight decision models now handle about a quarter of all Gateway requests, while Anthropic captured roughly twice the share of newly formed teams compared to OpenAI.

What changed in October 2026

  • Decision models reached about 26% of all AI Gateway requests in October, hitting "one in four" requests within six weeks of their mid-September 2026 launch.
  • 15 decision models from 8 vendors are now available on AI Gateway, costing roughly a seventieth as much per token as a language model.
  • Close to a third of all paid teams used a decision model in October; among teams processing over 10 billion tokens per month, seven in ten did.
  • Voice model adoption rose to about 11% of paid teams in October, up from 4% in July, a 161% increase.
  • New teams picked Anthropic over OpenAI by nearly two to one in October.
  • AI Gateway now routes more than 100 trillion tokens per month.

What it means for AI builders

The report is one of the largest third-party datasets on how production AI apps actually pick and route models, giving builders concrete usage signals to weigh when deciding what to build on. The shift toward cheap decision models, priced at roughly a seventieth of a language model's per-token cost, points to a growing pattern of using small models to make routing or classification calls before handing work to a larger model. Anthropic's lead in new-team adoption suggests where developer momentum is currently heading. The report also found that more than half of October's language-model requests went to models released within the prior two months, and that three in four teams abandon a given model within one month of its peak usage, a reminder that production AI stacks are being rebuilt on short cycles.


Sources

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.

Vercel AI Gateway report: decision models hit 26% of requests | vibecoding.inc