mammoth-1 / elephant-1

Base Labs (Baseten) full fine-tunes of Qwen3.5-27B and -9B, MIT weights, with a joey-1 2B sibling (text + image, provider-advertised $0.01/M and 20 ms p50) now completing the size ladder. The architectural move: block-bidirectional attention on Qwen3.5's full-attention layers (the GDN layers stay causal), so each question encodes knowing its own options while the state prefix stays cacheable across questions — worth +3.8 Decision Index points over the causal mask at fixed data, and the report notes Perplexity's Decider v1.1 applies a similar change. Trained in a week on 6.21M examples (cross-entropy plus a Brier term, and uncertainty-aware abstention rows that add an "I don't know" option and drop the correct answer), with the full eight-mix data evolution published: 45.5 → 74.0 on the public split. mammoth-1 led the Jev Decision Index at 75.34 on the Oct 29 draw (#1 of 247) and still leads JevBench v1.8's open capability view at 85.2, but the Nov 6 draw (edition 0.4.1, private-test reweighting) drops it to #13 of 274 (68.44) under REAX 1.3 — an ensemble whose Q-part blends a mammoth descendant; elephant-1 ties d1-ultra at 65 on Artificial Analysis's Decision Index.