Baseten: Mammoth 1 (free)

Baseten: Mammoth 1 (free) is a decision model from Baseten, published under the model id baseten/mammoth-1:free. It comes from Base Labs, Baseten's research team, and is built on Qwen3.5-27B. Its main category is decisions, and that single category describes what the model is for: rather than producing open ended prose, it is designed to answer structured questions about a given state.

The model takes text as input and returns decisions as output. The provider description frames the input as a state, which can be text or JSON, together with typed questions. The supported question types are yes/no, multiple choice, and ordinal. The context window is 16,384 tokens, and the maximum output is 14,745 tokens. There are no supported request parameters listed beyond the basics of an OpenAI-compatible call, so the model is configured through the prompt content rather than through sampling or tool options.

Practical use follows directly from that shape. You can pass a JSON state, such as a record, a configuration, or a set of extracted fields, and ask a yes/no question to get a binary decision. Multiple choice questions suit routing or classification where the answer must be one of a fixed set of options. Ordinal questions fit cases where the answer sits on a scale, such as severity or priority levels. Because the state can be plain text, the same model can also judge a passage, a support ticket, or a log excerpt. The 16,384 token context sets the ceiling on how much state and how many questions you can send in one call.

To call it through api14, use the OpenAI-compatible chat completions endpoint with the model id baseten/mammoth-1:free. Send your state and typed questions as the message content, and read the decision from the response. No extra request parameters are listed as supported, so keep the request to the standard fields and put the decision logic in the prompt itself. If your client library expects a parameters object, leaving it empty is the safe choice.

The main constraint to keep in mind is the context window of 16,384 tokens, which is modest if you need to include long documents alongside your questions. The output modality is decisions, so this is not the model to pick when you need generated prose, summaries, or code. If your task is open ended writing or general chat rather than answering typed questions about a state, another model will fit better.