LLM Stats (OCRBench-V2 (en)): leaderboard
Extended English OCR benchmark with harder recognition scenarios including dense tables, mathematical formulas, and degraded document images.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 32B (Thinking) | 68.4 | #294 (Qwen 3 VL 32B) |
| 2 | Nemotron 3.5 Super VL (BF16) | 68.12 | #165 |
| 3 | Qwen 3 VL 32B Instruct | 67.4 | #283 |
| 4 | Qwen 3 VL 235B A22B Instruct | 67.1 | #273 |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 66.8 | #238 (Qwen 3 VL 235B A22B) |
| 6 | Qwen 3 VL 8B Instruct | 65.4 | #420 |
| 7 | Qwen 3 VL 8B (Thinking) | 63.9 | |
| 8 | Qwen 3 VL 4B Instruct | 63.7 | #515 |
| 9 | Qwen 3 VL 30B A3B Instruct | 63.2 | #384 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 62.6 | #351 (Qwen 3 VL 30B A3B) |
| 11 | Qwen 3 VL 4B (Thinking) | 61.8 | #480 (Qwen 3 VL 4B) |
| 12 | Qwen 2.5 VL 72B Instruct | 61.5 | #377 |
| 13 | Qwen 2.5 VL 32B Instruct | 57.2 | #453 |
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-v2-en · How It Works · Data refreshed daily, snapshot 2026-10-20.