Vercel Data: Decision Models Hit 26% of Requests; Anthropic Wins New Teams 2:1 Over OpenAI

Cloud development platform Vercel has released its AI Gateway Production Index for November 2026, revealing a structural shift in enterprise AI usage. The data shows that "Decision Models"—specialized models designed for classification, scoring, or selection rather than text generation—accounted for 26% of all gateway requests in October. Despite handling a quarter of the traffic, these models represented only 0.53% of total spending, as their cost per request is approximately one-seventieth that of large language models.

This trend indicates that production AI applications are evolving from simple text generation to a separation of "generation and judgment." Currently, 15 decision models from eight different vendors are available on the gateway. TypeSafe AI's Jev model captured the vast majority of this volume. Notably, high-volume teams (processing over 10 billion tokens per month) showed a 70% adoption rate for decision models, compared to just one-fifth among lower-volume teams.

Regarding competition among leading labs, Anthropic and OpenAI have reached parity in terms of paid team coverage. However, Anthropic was the first choice for nearly twice as many new teams in October as OpenAI. Meanwhile, OpenAI's share of gateway spending increased by two-thirds since December, while Anthropic's share halved to about 32 cents per dollar. This suggests that while Anthropic is strong in acquiring new customers, OpenAI continues to grow its share of high-value workloads among existing large clients.

Additionally, voice model adoption is rising rapidly. About 11% of paid teams used voice models in October, up from 4% in July, representing a 161% increase. Most teams construct voice interactions by combining speech-to-text, language model response, and text-to-speech steps, though the use of end-to-end integrated models is growing.