mammoth-1, a restricted-access 27B, is the new #1 on Decision Index 0.4
publicado (OK do Fabricio 20/10 BRT).
Update 22 Oct 2026. On 21 Oct Base Labs opened both models under MIT and Baseten launched them with a hosted API, so they now have a catalog page. See the follow-up.
The Decision Index moved to edition 0.4 on 20 October 2026: the Full score now comes from the private parts only (60% same skills, 40% new domains), and 0.3.2 stays as an archive. Its new first place is mammoth-1 (27B) from baselabs, at Full 71.24, 1.82 points above Perplexity Decider v1.2 (69.42), so outside the 0.9 tie band. It is also first on the public part (75.18). For reference, Jev 1.13 is 61.26 (#34) and Jev 1.14 is 62.22 (#28).
What it is. The model card describes a full fine-tune of Qwen3.5-27B that answers every question of a request with a probability per option from one shared prefill, without generating text. A 9B sibling, elephant-1, is at 62.65 (#25).
Why it is not in the catalog. Both repositories need manual approval and say “All rights reserved. Access is granted for Decision Index evaluation only.” There is no API, no licence for use and no documentation beyond the kit command, so they fail our availability and documentation criteria. We list them as adjacent and will revisit if baselabs opens them.
Also new in 0.4: d3-turbo (news) at 69.04 and Drex Max (news) at 68.51.