Leaderboard
OCR for Danish, Norwegian and Swedish.
22 models · 10 datasets
Request a modelMetric
Word error rate (WER) counts word substitutions, deletions and insertions relative to the reference text.
Character error rate (CER) measures the same errors at the character level.
Lower is better. Extra text can push either score above 100%.
Read more about metrics →Mean: average error rate across tasks, with equal weight for each task. Lower is better.
Win rate: estimated chance of beating another model, fitted from task comparisons using Bradley–Terry. Higher is better.
Borda rank: average rank across tasks; ties share the average position. Lower is better.
Overall scores require results on every task. Individual task cells always show the selected error rate.
Read more about aggregation → Loading scores… About this benchmark ↗
Lower error, fewer parameters.
Pareto frontierOther models