Baseten's Base Labs open-sources decision models mammoth-1 (27B) and elephant-1 (9B) under MIT: mammoth-1 tops the Jev Decision Index at 71.24, elephant-1 ties Liquid's closed d1-ultra on Artificial Analysis

Why it matters

【Key points】

Base Labs, Baseten's research team, released open decision models mammoth-1 (27B) and elephant-1 (9B) on Oct 21 (17:14 UTC), MIT-licensed. Decision models return a probability per option in one forward pass instead of text, for routing, classification and scoring. Both are full fine-tunes of Qwen3.5 (text-only). mammoth-1 is #1 of 208 on the Jev Decision Index 0.4 at 71.24, ahead of Perplexity Decider v1.2 (69.42) and Jev 1.14 (+9.0). elephant-1 (62.65) ties Liquid's closed d1-ultra at 65 on Artificial Analysis's official index; mammoth-1's 65.9 there is a self-run replica. Free on Baseten and Vercel AI Gateway through Oct 27.

【Verifiable】

https://labs.baseten.co/articles/mammoth-1-and-elephant-1-frontier-decision-models

https://huggingface.co/baselabs/mammoth-1 , https://huggingface.co/baselabs/elephant-1

https://huggingface.co/spaces/multimodalart/jev-decision-index

【Our read】

1. Decision models only became a category with Jev in September. OpenAI, Liquid, Cloudflare and Perplexity followed, and TypeSafe raised $870M. Base Labs topped the board in about a week on an open base, mostly through data work, so the barrier here is data.

2. It edges out the best open rival, pplx-decider v1.2 (same size and Qwen base), under a looser license. elephant-1 matches a closed API model at 9B.

3. Closed decision-model vendors now have to compete on price and modality. Baseten joins the inference clouds that train their own models.

4. Caveats: the 1.8-point lead is small, the board is community-run, and both models are text-only.