BaFCo - Key Information Extraction (NES): leaderboard
BaFCo (2026) is a Bangla form-comprehension benchmark built from 200 multi-page Bangladeshi government forms. Key information extraction on Bangla forms scored by normalized edit similarity between extracted and gold values, in percent; higher is better.
Metric: Normalized edit similarity between extracted and gold values, one minus normalized Levenshtein distance (%; value extraction from Bangla government forms: given each key, return its value from the form image, formulated as visual question answering; 1,926 key-value pairs on 186 pages, Bangla subset; zero-shot prompt). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 86.6 | #137 |
| 2 | Qwen 3.6 Plus | 80.4 | #222 |
| 3 | GPT-5.2 | 79.6 | #176 |
| 4 | Claude Opus 4.6 | 70.4 | #102 |
| 5 | Kimi K2.5 | 70 | #256 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=bafco-key-information-extraction-nes · How It Works · Data refreshed daily, snapshot 2026-11-13.