Model rankings

426systems clear the coverage gate (≥4 benchmark families across ≥3 categories). Systems with less evidence appear on their model pages with full per-benchmark scores — absence of coverage is shown as absence, never as a low score. Reasoning-effort variants rank as separate systems; “best per model” (on by default) collapses them to each model’s best-ranked variant. Coverage dots scale against ~16 index families — hover for exact counts and confidence.