Decision models reach 26% of Vercel AI Gateway requests
Decision models accounted for 26% of requests routed through Vercel AI Gateway in October, while Anthropic became the first choice for nearly twice as many new teams as OpenAI. Vercel’s November Production Index also reports accelerating voice adoption and rapid movement to newly released language models.
- Decision models handled 26% of AI Gateway requests in October.
- Decision models represented 0.53% of total gateway spend.
- Anthropic attracted nearly twice as many new teams as OpenAI.
- Voice model adoption rose from 4% in July to 11% in October.
- More than half of language model requests went to models less than 60 days old.
Decision models add a lower cost option
Vercel said decision models have quickly become a significant part of production AI traffic. These models classify, score, or choose between options rather than generate text, and the company says they cost about a seventieth as much per request as a language model.
Decision models arrived on AI Gateway in mid-September. There are now 15 available from eight vendors. Nearly a third of paid teams used one in October, while 11% used a decision model without sending a request to a text-generating model.
The models received about one in four gateway requests but represented only 0.53% of total spend because they use fewer tokens and produce very short answers. Almost all of that traffic went to Jev from TypeSafe AI, which received nearly four times Anthropic’s total requests. For every $1 spent on Jev, teams spent $64 on Anthropic models.
Adoption was highest among the biggest teams. Seven in ten teams processing more than 10 billion tokens a month used a decision model, compared with just over one in five teams processing fewer than 10 million tokens.
Voice use rises as teams expand existing apps
About 11% of paid teams used a voice model in October, up from 4% in July. Vercel describes that as an increase of 161%. Among teams that paid in both months, voice adoption rose from 5% to 14%.
A typical spoken AI interaction combines speech to text, a language model that writes the reply, and text to speech. Use of speech to text models nearly tripled since July, while text to speech use more than tripled. In July, about a quarter of voice teams used an all in one model and one in six assembled the individual components. By October, those proportions had reversed.
The report says voice is usually being added as another way for users to interact with an existing app or agent. The share of teams sending at least 100 voice requests a month grew fivefold, and six in ten of September’s voice teams returned in October.
Anthropic gains new teams as model switching accelerates
OpenAI and Anthropic reached nearly the same share of paid teams in October. OpenAI was used by 55 of every 100 paid teams, compared with 54 for Anthropic. Among teams that joined the gateway in October, Anthropic was the first choice for nearly twice as many as OpenAI.
OpenAI nevertheless increased its share of gateway spend to nearly 15 cents of every dollar, up from about 9 cents in December. Anthropic remained ahead by spend, although its share fell to about 32 cents from about 64 cents over the same period, according to Vercel.
More than half of October’s language model requests went to models less than 60 days old. Within a model family, a new version arrives every 61 days at the median, and a typical model reaches peak usage within a month of launch. Four weeks later, three in four teams using it at its peak are no longer using it.
About half of teams processing at least 10 million tokens a month changed their most used model in October. Four in five of those teams moved to a more recent release, while six in ten moved to a different lab. Vercel says its report uses anonymized, aggregate traffic routed through AI Gateway through October 2026, with spend estimated from published list prices.
