Baseten open-sources mammoth-1, a top-ranked decision model
Why it matters
Base Labs, Baseten's research lab, released two open-weight decision models on Oct 21: mammoth-1 (27B) and elephant-1 (9B). A decision model takes an input, such as a support ticket, plus a set of questions and returns a probability for every answer option instead of writing text, so nothing is generated, there is no output to parse, and an option can be chosen or escalated only when it is likely enough. Base Labs reports mammoth-1 ranks #1 of 208 models on the Jev Decision Index 0.4 with 71.24, ahead of Perplexity Decider v1.2 (69.42); elephant-1 is the best model under 15B on that index (62.65) and level with Liquid AI's d1-ultra (65) on the Artificial Analysis Decision Index. Both were fully fine-tuned from Qwen3.5 for one epoch on 6.21M examples, 1.00M of them on safety and abstention, after the team found decision models were terrible at knowing when not to decide. The weights are on Hugging Face under an MIT license, and both models are free on Baseten's Model APIs until Oct 27, 11:59 PM PT.
Read the source: Base Labs (Baseten) (opens in a new tab)
All hits from 23 Oct 2026More Open ModelsGet the daily hits by email