OpenVLM OCRBench - Scene Text-centric VQA: leaderboard
OCRBench Scene Text-centric VQA from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): answering questions about text in natural scenes. This part holds 200 of OCRBench's 1,000 questions; the board shows correct answers as a percentage of those 200, higher is better.
Metric: Correct Answers (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 285 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 94.5 | #351 |
| 2 | InternVL3-78B | 93 | #361 |
| 3 | MiniCPM-V-2.6 | 93 | #827 |
| 4 | Gemini 2.5 Pro | 92.5 | #159 |
| 5 | Qwen 2 VL 7B | 92 | #743 |
| 6 | InternVL2.5-78B | 91.5 | #434 |
| 7 | GPT-4.1 (2025-04-14) | 91 | #257 |
| 8 | Qwen 2 VL 72B | 91 | #450 |
| 9 | Qwen 2 VL 2B | 91 | #1191 |
| 10 | Llama 3.2 11B Instruct | 91 | #1108 |
| 11 | Qwen 2.5 VL 7B | 90.5 | #641 |
| 12 | GPT-4.1 Mini | 90 | #363 |
| 13 | InternVL2.5-2B | 90 | #1227 |
| 14 | Llama 3.2 90B Vision Instruct | 90 | #621 |
| 15 | GPT-4o ChatGPT | 90 |
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-scene-text-centric-vqa · How It Works · Data refreshed daily, snapshot 2026-11-05.