Observed arrival · 2026-11-21
BB Index: a model leaderboard you can reweight
BB Index combines public model benchmarks into rankings, then lets users adjust benchmark weights using their own comparisons.
- For
- Researchers comparing language-model benchmark results
- Worth noticing
- Its composite ranking requires rankable measurements in all six weighted domains; missing data is neither predicted nor filled with zero.
Field notes
The methodology groups related evaluations into benchmark families to limit duplicate weighting, compares only measurements shared by two models, and preserves genuine zero scores while treating missing results as missing. Its composite ranking requires rankable results in all six domains; models missing a domain are excluded rather than imputed. The page reports that user calibration changes shared evaluation weights, not the scores of models named in a comparison.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue