KhatianDoc - Symbol Recognition CER: leaderboard
KhatianDoc (2026) is a human-verified benchmark on real Bangladeshi RS Khatian land records in the Ana-Ganda base-16 fractional numeral system, evaluated zero-shot. This board is symbol recognition scored by character error rate over 95 numeral strings, in percent, lower is better.
Metric: Character error rate (%; symbol recognition of Ana-Ganda numerals over 95 strings, zero-shot; lower is better). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 76.89 | #982 |
| 2 | Qwen 2.5 VL 72B Instruct | 80.91 | #845 |
| 3 | Gemma 4 26B A4B (IT) | 84.7 | #545 |
| 4 | Llama 4 Scout | 90.91 | #1787 |
| 5 | Gemini 2.5 Flash Lite | 93.29 | #958 |
| 6 | GPT-4o Mini | 96.56 | #1627 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=khatiandoc-symbol-recognition-cer · How It Works · Data refreshed daily, snapshot 2026-11-12.