We use cookies. Cookies and browser storage keep you signed in and remember your preferences. With your permission we also measure how the site is used. You can any time. See our cookie policy.
baseten/elephant-1Runs on your balance at the prices above. New accounts start with free usage.
Settings
Baseten: Elephant 1 is a decision model from Baseten, developed by Base Labs, the company's research team. It is built on Qwen3.5-9B and has 9B parameters. Rather than generating free-form conversation, the model is designed to return decisions, which places it in the decisions category on api14. If you are building a system that needs a model to commit to a choice rather than produce prose, this is the kind of model to evaluate.
The model accepts text input and produces decisions as output. Its context window is 16,384 tokens, and the maximum output is 14,745 tokens. Inputs can take the form of a state, provided as text or JSON, together with typed questions. The supported question types are yes/no, multiple choice, and ordinal. Because the output modality is decisions, you should expect structured answers to the questions you pose rather than narrative text, and design your parsing and downstream logic around that.
The practical use cases follow directly from that design. A state plus a yes/no question suits approval gates, eligibility checks and policy screening, where a pipeline needs a clear verdict on a record or request. Multiple choice questions fit routing and classification tasks, such as sending an incoming ticket to the right queue or picking a category for a document. Ordinal questions are useful when the answer sits on a scale, for example grading severity or priority from a supplied state. Supplying the state as JSON fits workflows where the input already comes from an API or database and you want to pass it through without reshaping it into prose. In each case the model reads the state and answers the typed question, which keeps the output narrow and easier to validate than free text.
You call the model through api14's OpenAI-compatible API using the model id "baseten/elephant-1". Because the interface follows the OpenAI format, existing OpenAI client libraries and request patterns carry over, with the model id swapped in. The model lists no supported request parameters beyond what the API accepts by default, so keep requests to the standard fields and avoid relying on optional tuning parameters such as temperature or top_p; check the live parameter list on the page before you build around any of them. The 16,384 token context window bounds how much state and how many questions you can send in a single request, so keep inputs within that budget.
The model is text in and decisions out, so it does not fit work that needs prose, summaries or other generated text; a general text generation model is the better choice there. The absence of supported parameters also means you cannot shape sampling behaviour through the request, which matters if you need varied or creative output. For decision tasks over a bounded state, with typed questions and a context that fits within 16,384 tokens, Elephant 1 is a focused option to test against your own workloads.
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api14.com/api/v1",
apiKey: process.env.API14_API_KEY,
});
const result = await client.post("/decisions", { body: {
"model": "baseten/elephant-1",
"state": "Customer: I was charged twice for my order and want my money back.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should handle this message?",
"criteria": {
"billing": "Payments, refunds, invoices",
"support": "Product questions and bugs",
"sales": "Pricing and new purchases"
}
},
"urgent": {
"type": "noul",
"instructions": "Does this need a reply today?",
"criteria": {
"true": "Money, outage or legal risk",
"false": "Can wait"
}
}
}
} });
console.log(result.answers);
This community has no topics yet. Open the first one about Baseten: Elephant 1.