OCR-Robust: leaderboard
Vision-language models (VLMs) have achieved strong performance on OCR-based benchmarks and increasingly focused on text-rich understanding, but their robustness under controlled visual degradation remains insufficiently understood. This gap is critical for OCR reasoning, where visual corruption can induce OCR errors and structural distortions, thereby introducing uncertainty into the reasoning tas
Metric: OCR1.0 Clean (self-reported). Source: benchmarklist.com. Saturation forecast: Around April 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 80.71 | #99 |
| 2 | Qwen 3 VL 8B (Thinking) | 79.67 | |
| 3 | Qwen 3 VL 4B (Thinking) | 78.63 | #480 (Qwen 3 VL 4B) |
| 4 | GPT-5.2 | 74.69 | #110 |
| 5 | Qwen 3 VL 32B Instruct | 72.82 | #283 |
| 6 | Llama 4 Scout | 67.84 | #656 |
| 7 | GPT-5.1 | 60.58 | #135 |
| 8 | Qwen 3 VL 8B Instruct | 54.15 | #420 |
| 9 | Qwen 3 VL 4B Instruct | 49.79 | #515 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ocr-robust · How It Works · Data refreshed daily, snapshot 2026-10-20.