Elephant-1

Base Labs' 9B open-weight decision model for classification, routing, and scoring.

Also known as elephant-1, baseten-elephant-9b

What it does

Elephant-1 is a 9 billion parameter decision model from Base Labs at Baseten, released with Mammoth-1 in a technical report dated October 21, 2026. It is a full fine-tune of Qwen3.5-9B. You give it a state, as text and optionally images, and a set of questions, and it returns a probability for each allowed answer instead of generating text. The model card's question types are yes/no, a choice among named options, and a score on ordered levels. Base Labs scores it at 62.65 on the Jev Decision Index 0.4 and calls it the best decision model under 15 billion parameters on that index. The model card lists an MIT license and a 16,384-token maximum sequence length, and says the Baseten API is free through October 27, 2026 at 11:59 PM PT, with zero data retention on by default.

Who is Elephant-1 for?

Developers who want a smaller open-weight model for routing, classification, or rubric scoring.

Tags

• open-weight
• routing
• classification