Benchmark · arXiv 2604.12978

GlotOCR-Bench Leaderboard

↗ View paper April 2026 · v1.0
Models evaluated
14
open-weight & API
Unicode scripts
158
1 high · 9 mid · 148 low
Best overall Acc@5
—
—
Low-resource ceiling
7.7%
148 scripts, best model
Filter —
Variant —
# Model ↕ Overall ↑ ▼ Macro CER ↓ ↕ High — Latin ↕ Mid — 9 scripts ↕ Low — 148 scripts ↕
Overall = avg(High, Mid, Low)
High = Latin
Mid = Arab · Cyrl · Deva · Hani · Jpan · Hang · Grek · Hebr · Thai
Low = 148 remaining scripts
Script — Variant — Sort —
0%
100% = Acc@5