OpenVLM OCR-VQA: leaderboard
OpenCompass OpenVLM OCR-VQA test-core results: question answering over book-cover images where models must read title, author, genre, and other visual text.
Metric: Accuracy (%). Source: ocr-vqa.github.io. Source: huggingface.co. 118 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2 VL 72B | 73.7 | #460 |
| 2 | Pixtral-12B | 64.7 | #808 |
| 3 | Aquila-VL-2B | 63.3 | #1238 |
| 4 | InternVL2-8B | 42.6 | #817 |
| 5 | InternVL3-8B | 38.9 | #620 |
| 6 | GPT-4.1 Nano | 36 | #731 |
| 7 | InternVL3-14B | 30.5 | #505 |
| 8 | Gemma 3 4B | 27.6 | #1089 |
| 9 | GPT-4.1 Mini | 26.3 | #355 |
| 10 | GPT-4.1 (2025-04-14) | 18.7 | #250 |
| 11 | Gemini 1.5 Flash | 18.5 | #618 |
| 12 | Gemma 3 12B | 17.7 | #671 |
| 13 | Gemma 3 27B | 13.5 | #607 |
| 14 | Gemini 1.0 Pro | 12.3 | #933 |
Interactive version: theaggregate.ai/benchmark?slug=openvlm-ocr-vqa · How It Works · Data refreshed daily, snapshot 2026-10-20.