OpenVLM OCRBench - Handwritten Mathematical Expression Recognition: leaderboard
OCRBench Handwritten Mathematical Expression Recognition from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): transcribing handwritten mathematical expressions. This part holds 100 of OCRBench's 1,000 questions; the board shows correct answers as a percentage of those 100, higher is better.
Metric: Correct Answers (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 285 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Moonshot v1 8K | 89 | #622 |
| 2 | InternVL3-78B | 85 | #361 |
| 3 | InternVL3-8B | 83 | #624 |
| 4 | InternVL3-38B | 80 | #413 |
| 5 | InternVL3-14B | 79 | #506 |
| 6 | InternVL2.5-78B | 78 | #434 |
| 7 | Qwen 2.5 VL 7B | 76 | #641 |
| 8 | InternVL2.5-2B | 76 | #1227 |
| 9 | Qwen 2 VL 72B | 73 | #450 |
| 10 | Aquila-VL-2B | 64 | #1216 |
| 11 | InternVL2-8B | 63 | #813 |
| 12 | Qwen 2 VL 7B | 59 | #743 |
| 13 | MiniCPM-V-2.6 | 48 | #827 |
| 14 | Qwen 2 VL 2B | 39 | #1191 |
| 15 | Gemini 2.5 Pro | 38 | #159 |
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-handwritten-mathematical-expression-recognition · How It Works · Data refreshed daily, snapshot 2026-11-05.