Baseten's Mammoth-1 and Elephant-1 decision models free on AI Gateway
Vercel has made Baseten's Mammoth-1 and Elephant-1 decision models available for free on AI Gateway for one week, with free access running through October 27. The models classify text, route requests and score content against criteria defined by the developer.
- Baseten's Mammoth-1 (27B) and Elephant-1 (9B) are free on Vercel AI Gateway through October 27.
- The decision models classify text, route requests and score content against developer-defined criteria.
- Per Vercel, Mammoth-1 is #1 on the Jev Decision Index and Elephant-1 is tied for #1 on the Artificial Analysis Decision Index.
- Both models support a 16k-token context window and return choices, probabilities and scores.
- One request can include Choice, Boolean and Score questions, and the application sets the probability threshold.
Two decision models, two sizes
Mammoth-1 and Elephant-1 are what Vercel calls decision models. According to the announcement, they classify text, route requests, and score content against criteria that developers define themselves. They are not described as general chat models, but as tools for making structured choices about an input.
The two models differ in size. Mammoth-1 is the larger option at 27B parameters, while Elephant-1 is the smaller 9B model. Both support a 16k-token context window and return choices, probabilities and scores.
Vercel also cites benchmark placements. According to the announcement, Mammoth-1 is #1 on the Jev Decision Index, and Elephant-1 is tied for #1 on the Artificial Analysis Decision Index. These are claims from the announcement, and the source gives no further detail on how the rankings were produced.
Several questions in one request
A key feature is that developers can ask several questions about the same input in a single request. Vercel gives three examples of question types:
- Choice: select a support team.
- Boolean: check whether a message contains a refund request.
- Score: rate how urgent a message is.
Because the models return probabilities, the application decides what threshold to use. Above that threshold, it can act automatically. Below it, a result can be sent for review.
The model names to use are baseten/mammoth-1 and baseten/elephant-1. Vercel's example uses the experimental decide function from the AI SDK. The sample passes the text "I was charged twice. Please refund the duplicate." as the state, along with a Boolean question named refund that asks whether the customer is requesting a refund. The result is then read from result.answers.refund. The announcement points to the decision models documentation for question types and further usage details.
What it means for developers
The free window lasts one week, so developers who want to try the models on tasks such as ticket routing, refund detection or urgency scoring can do so on AI Gateway without model costs until October 27. The announcement does not state pricing after that date.
AI Gateway provides a unified API for accessing models and tracking usage and cost. According to Vercel, decision calls appear in logs and count toward budgets alongside other model requests, so teams can monitor them the same way as their other model usage. Vercel also lists all decision models available on AI Gateway on a separate page.
