02 / IT · Analysis
Decision models reshape AI Gateway requests and spending
Decision models account for sharply different shares of requests and spending, while new teams’ provider choices and spending shares also diverge.

Vercel reported that decision models account for 26% of AI Gateway requests but just 0.53% of spending in its November report. For developers choosing models for their apps, the gap shows why request volume and cost need separate consideration.
Vercel describes decision models as models that classify, score or choose between options instead of generating text. Its report separates generation from judgement, showing how models that receive many requests can differ from those that attract the most spending.
The report also puts Anthropic ahead among new teams, while OpenAI’s share of spending is rising. Provider reach and spending share measure different aspects of adoption. That distinction matters both for decision models and for competition between language-model providers.
Decision models account for 0.53% of spend
| Decision models | Request share | 26 | |
|---|---|---|---|
| Spend share | 0.53 |
Decision models change the request mix
According to Vercel, decision models produce short answers and cost less per token. Most decision-model requests went to Jev from TypeSafe AI. In the report, higher request volume does not translate into a matching share of spending.
Vercel also counted paid teams that used decision models without sending requests to text-generating models. Adoption is higher among teams processing more tokens. The report presents this as a different pattern from adding capabilities such as voice to a text-generating model.
The report’s methodology specifies anonymised, aggregate traffic routed through AI Gateway. It defines paid teams as those with a paid request during the month. Its team-adoption findings therefore describe AI Gateway’s paid teams.
New-team choices and spending diverge
Vercel defines a new team’s first-choice provider as the company receiving the most language-model requests from that team in its first month. On that measure, nearly twice as many new teams chose Anthropic as OpenAI. Across all paid teams, the providers’ reach has become similar.
According to Vercel, most of the highest-spending teams use models from both companies. Anthropic leads in spending share, while OpenAI has gained ground. New-team preferences, use of both providers among large spenders and overall spending shares are separate measures.
The report states that spending is estimated at published list prices and that actual bills may differ. Its spending shares therefore describe estimated costs. For teams assessing model choices, request volume, provider adoption and estimated spending each offer a different part of the picture.