KhatianDoc - Field Extraction Row F1: leaderboard
KhatianDoc (2026) is a human-verified benchmark on 107 real Bangladeshi RS Khatian land records written in the Ana-Ganda-Kora-Kranti-Til base-16 fractional numeral system, evaluating six multimodal LLMs zero-shot through OpenRouter. This board is structured field extraction scored by row-level F1 over 261 rows, in percent, higher is better; it is the papers metric of record for extraction.
Metric: Row-level F1 (%; structured field extraction over 261 rows in 107 records, zero-shot; the metric of record for extraction). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 26B A4B (IT) | 25.84 | #545 |
| 2 | Qwen 3 VL 8B Instruct | 17.48 | #982 |
| 3 | Qwen 2.5 VL 72B Instruct | 11.97 | #845 |
| 4 | Llama 4 Scout | 6.43 | #1787 |
| 5 | Gemini 2.5 Flash Lite | 5.16 | #958 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=khatiandoc-field-extraction-row-f1 · How It Works · Data refreshed daily, snapshot 2026-11-12.