Devanagari OCR Stress Test - Clean Render: leaderboard
Devanagari OCR stress test (Aditya Pratap Singh, 2026) comparing classical OCR engines, open and specialised OCR vision-language models and frontier multimodal APIs on Hindi text. This set is the first 100 Hindi sentences of the FLORES test set rendered cleanly in five Devanagari fonts. Transcriptions are NFC-normalised and scored with chrF++ (character n-gram F-score with word bigrams, 0-100); higher is better.
Metric: chrF++ (0-100), 100 FLORES Hindi sentences rendered in five Devanagari fonts, clean images, NFC-normalised. Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 98 | #75 |
| 2 | Gemini 2.5 Flash | 96.5 | #452 |
| 3 | GPT-5.5 | 96.5 | #48 |
| 4 | Qwen 3 VL 8B Instruct | 95.9 | #982 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=devanagari-ocr-stress-test-clean-render · How It Works · Data refreshed daily, snapshot 2026-11-12.