The source is labeled as an October 2026 checkpoint. Rather than generating text as the primary output, the models take a JSON state and questions, then return probability distributions for choices, yes/no answers, or scores.
Base Labs says mammoth-1 ranks first among 208 models on Jev Decision Index 0.4, with a score of 71.24. It reports 65.9 for mammoth-1 and 65.2 for elephant-1 on its replication of the Artificial Analysis Decision Index. The company says both models were post-trained from Qwen3.5-9B and Qwen3.5-27B, using block-bidirectional attention for full-attention layers and a data-processing recipe.
The article describes confidence values and constrained decision outputs for uses such as ticket routing. It does not establish independent benchmark verification, model weights or API availability, licensing terms, pricing, or deployment limits.