elephant-1 9B

elephant-1 is the small brother of mammoth-1, and proof the decision-model category rewards small: 9B parameters, one forward pass, a confidence number per option. The analogy is the search dog rather than the reference library: it is not trying to know everything, it is trying to pick the right shelf fast. It exists because a 9B model that beats generalist giants on routing work costs almost nothing to run. Reach for it as the tier of decision models that fits on a laptop alongside your main model. Honest status: weights on HuggingFace (MIT); claimed best under 15B on the Jev index, vendor-stated, verifiable at home.