OpenVLM OCRBench - Key Information Extraction: leaderboard
OCRBench Key Information Extraction from the OpenCompass OpenVLM Leaderboard (VLMEvalKit): extracting named fields from receipts, forms and product labels. 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: Around February 2027. 285 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2 VL 72B | 94 | #450 |
| 2 | Gemini 2.5 Pro | 93.5 | #159 |
| 3 | Qwen 2 VL 7B | 93.5 | #743 |
| 4 | InternVL3-14B | 92.5 | #506 |
| 5 | InternVL3-78B | 92 | #361 |
| 6 | InternVL3-8B | 91 | #624 |
| 7 | InternVL3-38B | 91 | #413 |
| 8 | Qwen 2.5 VL 7B | 91 | #641 |
| 9 | GPT-4.1 Mini | 90.5 | #363 |
| 10 | MiniCPM-V-2.6 | 90.5 | #827 |
| 11 | Qwen 2 VL 2B | 90.5 | #1191 |
| 12 | GPT-5 Mini (2025-08-07) | 90.5 | #175 |
| 13 | InternVL2.5-78B | 89.5 | #434 |
| 14 | Claude 3.5 Sonnet (20241022) | 88 | #302 |
| 15 | Gemini 2.0 Flash | 88 | #351 |
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-key-information-extraction · How It Works · Data refreshed daily, snapshot 2026-11-05.