LLM Stats (OCRBench): leaderboard
1,000 images testing OCR capabilities across text recognition, scene text, handwriting, document layout, and formula recognition tasks.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Estimated already saturated. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 92.3 | #144 |
| 2 | Qwen 3.5 122B A10B | 92.1 | #173 |
| 3 | Qwen 3 VL 235B A22B Instruct | 92 | #273 |
| 4 | Qwen 3.5 35B A3B | 91 | #256 |
| 5 | Qwen 3 VL 30B A3B Instruct | 90.3 | #384 |
| 6 | Qwen 3 VL 8B Instruct | 89.6 | #420 |
| 7 | Qwen 3 VL 32B Instruct | 89.5 | #283 |
| 8 | Qwen 3.5 27B | 89.4 | #187 |
| 9 | Qwen 3.6 27B | 89.4 | #170 |
| 10 | Qwen 2.5 VL 72B Instruct | 88.5 | #377 |
| 11 | Qwen 3 VL 4B Instruct | 88.1 | #515 |
| 12 | Qwen 3 VL 235B A22B (Thinking) | 87.5 | #238 (Qwen 3 VL 235B A22B) |
| 13 | Qwen 2.5 VL 7B Instruct | 86.4 | #654 |
| 14 | Qwen 3 VL 32B (Thinking) | 85.5 | #294 (Qwen 3 VL 32B) |
| 15 | Phi-4 Multimodal Instruct | 84.4 | #906 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-ocrbench · How It Works · Data refreshed daily, snapshot 2026-10-20.