OCRBench v2
A large-scale bilingual (English/Chinese) text-centric benchmark of ~10,000 human-verified QA pairs across 31 scenarios that evaluates large multimodal models on visual text localization, recognition, parsing, and reasoning.
What is OCRBench v2?
A large-scale bilingual (English/Chinese) text-centric benchmark of ~10,000 human-verified QA pairs across 31 scenarios that evaluates large multimodal models on visual text localization, recognition, parsing, and reasoning. evals.report tracks reported OCRBench v2 scores with the model, source, status, date, and run caveats attached — official leaderboard scores, vendor-reported launches, and clearly labeled community runs.
Top reported OCRBench v2 score: Gemini 3 Pro — 63.4 (accuracy).
| Model | Lab | Score↓ | Source model | Status | Date | |
|---|---|---|---|---|---|---|
| Gemini 3 Pro | Google DeepMind | 63.4 | — | Verified | Nov 18, 2025 | Details |
| Gemini 2.5 Pro | Google DeepMind | 59.3 | — | Verified | Mar 25, 2025 | Details |
| GPT-5 | OpenAI | 55.5 | — | Verified | Aug 7, 2025 | Details |
| Gemini 1.5 Pro | Google DeepMind | 51.6 | — | Verified | Feb 15, 2024 | Details |
| GPT-5.2 | OpenAI | 50.5 | — | Verified | Dec 11, 2025 | Details |
| Claude Opus 4.6 | Anthropic | 48.4 | — | Verified | Feb 5, 2026 | Details |
| GPT-4o | OpenAI | 47.6 | — | Verified | May 13, 2024 | Details |
| Claude 3.5 Sonnet | Anthropic | 47.5 | — | Verified | Jun 20, 2024 | Details |
| Grok 4 | xAI | 45.0 | — | Verified | Jul 9, 2025 | Details |
| Claude Sonnet 4 | Anthropic | 42.4 | — | Verified | May 22, 2025 | Details |
Each row reports the model’s accuracy on OCRBench v2. Click a row for the full run context.