Baseten's Mammoth-1 and Elephant-1 decision models free on Vercel AI Gateway through October 27

Vercel has opened free access to two decision models from Baseten, Mammoth-1 and Elephant-1, through its AI Gateway. According to the Vercel blog, the promotion lasts one week, with free usage ending on October 27.

What decision models do

The two models are built for structured judgement calls rather than open-ended conversation. As Vercel describes them, they classify text, route incoming requests, and score content against criteria a developer defines. Mammoth-1 is the larger of the pair at 27 billion parameters, while Elephant-1 is a 9-billion-parameter option suited to lighter workloads.

Vercel reports that Mammoth-1 currently ranks first on the Jev Decision Index, and that Elephant-1 is tied for first on the Artificial Analysis Decision Index. Both models accept a 16,000-token context window and return choices, probabilities, and scores as their output.

Several questions per request

A single call can carry multiple questions about the same input, which reduces round trips for pipelines that need more than one signal from a piece of text. Vercel's examples include:

• a Choice question that picks which support team should handle a ticket,
• a Boolean question that determines whether a message contains a refund request, and
• a Score question that rates how urgent an item is.

Threshold logic stays with the application: developers decide at what probability the system acts automatically and when a result is instead sent for human review.

Access through AI Gateway

The models are reachable using the names baseten/mammoth-1 and baseten/elephant-1 through AI Gateway's unified API. The gateway already gives developers one interface for calling many models and tracking usage and spend, and Vercel notes that decision calls appear in the same logs and count toward the same budgets as any other model request. Vercel points developers to its decision models documentation for the full set of question types and additional usage details.

Why it matters

Routing, classification, and scoring are among the most common jobs teams hand to language models, yet they are typically done by prompting a general-purpose chat model and parsing its prose. Dedicated decision models return choices, probabilities, and scores directly, which removes the parsing step and makes thresholds explicit — useful both for full automation and for deciding when a human should look at a case.

The free week is a low-risk window to benchmark the two sizes against an existing routing setup, including the difference between the 27B and 9B variants on real workloads. Because the calls flow through AI Gateway, teams can also evaluate them under the logging, budgeting, and usage tracking they already use for other models, rather than standing up separate billing and observability for a new provider.

The offer runs through October 27; Vercel's announcement does not specify what pricing applies once it ends.