Tabular benchmarks
Every benchmark the Tabular index tracks, with each core benchmark's share of the composite and how many models it covers.
19 benchmarks tracked.
- BeyondArena Elo (IID+temporal+grouped, all)
- BeyondArena Elo (grouped splits)
- BeyondArena Elo (IID splits)
- BeyondArena Elo (temporal splits)
- MulTaBench (40 multimodal tables)
- ScoringBench CRPS average rank
- ScoringBench R2 average rank
- STRABLE (108 string-containing tables)
- TabArena Average Rank
- TabArena Elo (IID, default config)
- TabArena Elo (classification tasks)
- TabArena Elo (medium datasets, 10K-100K train rows)
- TabArena Elo (regression tasks)
- TabArena Elo (small datasets, <=10K train rows)
- TabArena Improvability (%)
- TabArena median predict time (s per 1K rows)
- TabArena Normalised Score
- TabArena median train time (s per 1K rows)
- TALENT average rank (300 datasets)