Atlas

Benchmarks

← All benchmarks

AA-Omniscience Public Index

Indexes · 2025-11-17

The AA-Omniscience Index computed on AA-Omniscience-Public, the 600-question open 10% subset of the original 6,000-question AA-Omniscience pool that Artificial Analysis released for self-evaluation. Answers are graded correct, incorrect, partial or not attempted, and the index is correct minus incorrect as a share of all questions on a -100 to 100 scale, so abstaining costs nothing while wrong answers are penalized.

Top models (higher is better)

ModelScore
LFM3-1.7B0.5
Qwen3.5-9B-28.0
LFM2.5-2.6B-33.2
Qwen3.5-4B-57.2
Loading Atlas data…