OpenVLM OCRBench: leaderboard
OpenCompass OpenVLM normalized OCRBench score, evaluating multimodal OCR and document-understanding ability on text recognition, scene text VQA, document VQA, key information extraction, and handwritten math expressions.
Metric: Score (normalized). Source: huggingface.co. Source: rank.opencompass.org.cn. Saturation forecast: Around January 2027. 285 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 90.8 | #352 |
| 2 | Qwen 2 VL 72B | 88.8 | #460 |
| 3 | Qwen 2.5 VL 7B | 88.8 | #643 |
| 4 | InternVL3-38B | 88.6 | #405 |
| 5 | InternVL3-8B | 88.4 | #620 |
| 6 | InternVL3-14B | 87.7 | #505 |
| 7 | Gemini 2.5 Pro | 86.2 | #152 |
| 8 | InternVL2.5-78B | 85.3 | #431 |
| 9 | MiniCPM-V-2.6 | 85.2 | #836 |
| 10 | Gemini 2.0 Flash | 85.1 | #342 |
| 11 | GPT-4.5 | 84.5 | #274 |
| 12 | Qwen 2 VL 7B | 84.3 | #750 |
| 13 | GPT-4.1 Mini | 84 | #355 |
| 14 | GPT-4.1 (2025-04-14) | 83.4 | #250 |
| 15 | Doubao-1.5-Pro | 82.9 | #287 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=openvlm-ocrbench · How It Works · Data refreshed daily, snapshot 2026-10-20.