index6 benchmarks

WeirdBench Intelligence Index

A single ranking across every WeirdBench benchmark. Raw scores are first converted into benchmark-local scores relative to the leader, so small raw gaps stay small while lower-is-better and higher-is-better benchmarks can still live in the same table.

Ranking

73 models
MethodologyHow index scoring works

Normalization

Higher-is-better benchmarks are scored relative to the leader. Lower-is-better benchmarks use the inverse ratio to the leader, with a safe fallback if scores cross zero.

Coverage

Models missing benchmarks are not dropped. Their average is multiplied by benchmark coverage ratio so partial coverage is visible and penalized.

Final Score

Final index = average normalized benchmark score × coverage ratio. Higher is better.