Baseten’s blog and changelog, dated October 21, 2026, introduce mammoth-1 and elephant-1 as decision models from Base Labs, Baseten’s internal research team. DigIntoAI’s news desk places the public announce around Beijing time October 22, 2026, 01:13, from Baseten’s X post 2112955524207825053 at 17:13:48Z on October 21.
Per Baseten, these are decision models that answer Choice, Score, and Boolean questions and return a probability distribution over options. Primary serving path is Baseten Model APIs via the System One endpoint POST https://inference.baseten.co/v1/systemone. Docs list model ids decision-models/baseten-elephant-9b-latest (context 16,384) and decision-models/baseten-mammoth-27b-latest (context 8,192). A second surface: Vercel AI Gateway also offers free access through October 27 as baseten/mammoth-1 and baseten/elephant-1. Vercel’s changelog claims both support a 16k-token context; Baseten’s docs list elephant at 16,384 and mammoth at 8,192.
Free window: free on Baseten Model APIs and on Vercel AI Gateway through October 27, 2026, 11:59 PM Pacific Time—that is Beijing time 2026-10-28 14:59. Both weights are open on Hugging Face (baselabs/mammoth-1, baselabs/elephant-1), post-trained from Qwen3.5-27B and Qwen3.5-9B.
On rankings, keep Baseten’s claims and the two indexes apart. On the Jev Decision Index 0.4, Baseten says mammoth-1 is #1 among 208 decision models with score 71.24, and elephant-1 is the leading 9B model (the blog also places elephant tied for #8 overall on that index). Separately, on the Artificial Analysis Decision Index, Baseten says elephant-1 is tied for #1 with d1-ultra, and cites scores of 65.9 for mammoth and 65.2 for elephant. Those standings are Baseten’s own citations of Jev and Artificial Analysis.
For how accurate or fast these decision models are, how billing works after the free window, and the leaderboard numbers, we have not tested yet. If you have used them, you can reach the site owner through the contact link in the footer. Related reading: OpenAI Decisions API Is Generally Available with GPT-6 Luna, MiniMax M3.1 Flash Open Weights Are on Hugging Face, and Muse Spark Open Weights Are Out. Here's Where You Can Run Them.