OCRBench v2
Multimodal · 2024-12-30
OCRBench v2 is a multimodal OCR benchmark covering recognition, extraction, parsing, understanding, and reasoning tasks across English and Chinese public/private splits.
Top models (higher is better)
| Model | Score |
|---|---|
| Interfaze 1 | 70.7 |
| Interfaze Beta | 70.7 |
| Gemini 3.7 Flash | 63.9 |
| Sonnet 5 | 59.2 |
| Grok 4.3 | 54.7 |
| GPT-5.4 Mini | 52.7 |