OCRBench — OCR benchmark; bucket mixes ~0-100 scale (median 87.3) w/ a raw-points outlier (865).
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen2-VL-72B-Instruct | Alibaba | 877 | 1 | 2026-08-24 |
| 2 | MiniMax VL 01 | MiniMax | 865 | 1 | 2026-06-05 |
| 3 | InternVL2.5-78B | OpenGVLab | 847 | 1 | 2026-06-05 |
| 4 | Gemini 2.0 Flash Exp | 846 | 1 | 2026-06-05 | |
| 5 | InternVL3_5-2B | OpenGVLab | 834 | 1 | 2026-08-09 |
| 6 | InternVL3-2B | OpenGVLab | 831 | 1 | 2026-08-09 |
| 7 | Qwen2.5 VL 3B | Alibaba | 824 | 1 | 2026-08-09 |
| 8 | LFM2-VL-3B | Liquid AI | 822 | 1 | 2026-08-09 |
| 9 | GPT-4o | OpenAI | 806 | 2 | 2026-06-12 |
| 10 | Llama 3.2 Instruct 90B (Vision) | Meta | 805 | 1 | 2026-06-05 |
| 11 | Gemini 1.5 Pro | 800 | 2 | 2026-06-12 | |
| 12 | Claude 3.5 Sonnet | Anthropic | 790 | 2 | 2026-06-12 |
| 13 | LFM2.5-VL-450M-Extract | Liquid AI | 684 | 1 | 2026-08-09 |
| 14 | LFM2-VL-450M | Liquid AI | 657 | 1 | 2026-08-09 |
| 15 | Kimi K2.5 | Moonshot | 92.3 | 2 | 2026-08-23 |
| 16 | Qwen3.5 122B A10B | Alibaba | 92.1 | 1 | 2026-08-23 |
| 17 | Qwen3 VL 235B A22B Instruct | Alibaba | 92 | 1 | 2026-08-23 |
| 18 | Qwen3.5 35B A3B | Alibaba | 91 | 1 | 2026-08-23 |
| 19 | Qwen3 VL 30B A3B Instruct | Alibaba | 90.3 | 1 | 2026-08-23 |
| 20 | Gemini 3 Pro | 90.3 | 1 | 2026-06-15 | |
| 21 | Qwen3 VL 8B Instruct | Alibaba | 89.6 | 1 | 2026-08-23 |
| 22 | Qwen3 VL 32B Instruct | Alibaba | 89.5 | 1 | 2026-08-23 |
| 23 | Qwen3.6 27B | Alibaba | 89.4 | 2 | 2026-08-24 |
| 24 | Qwen3.5 27B | Alibaba | 89.4 | 2 | 2026-08-24 |
| 25 | Step3 VL 10B | StepFun | 89 | 2 | 2026-06-05 |
| 26 | Qwen2.5 VL 72B | Alibaba | 88.5 | 1 | 2026-08-23 |
| 27 | Qwen3 VL 4B Instruct | Alibaba | 88.1 | 1 | 2026-08-23 |
| 28 | Qwen3 VL 235B A22B Reasoning | Alibaba | 87.5 | 3 | 2026-08-23 |
| 29 | Claude Opus 4.5 | Anthropic | 86.5 | 1 | 2026-06-15 |
| 30 | GLM-4.6V (106B-A12B) | Z.ai | 86.2 | 2 | 2026-06-05 |
| 31 | Gemma 4 31B | 86.1 | 1 | 2026-08-24 | |
| 32 | GLM-4.6V-Flash (9B) | Z.ai | 86 | 2 | 2026-06-05 |
| 33 | Gemini 2.5 Pro | 85.9 | 2 | 2026-06-05 | |
| 34 | Qwen3 VL 32B Reasoning | Alibaba | 85.5 | 1 | 2026-08-23 |
| 35 | MiMo VL RL 2508 (7B) | Xiaomi | 85.4 | 2 | 2026-06-05 |
| 36 | Phi 4 Multimodal Instruct | Microsoft | 84.4 | 1 | 2026-08-23 |
| 37 | LFM2.5-VL-3B | Liquid AI | 84.2 | 1 | 2026-08-23 |
| 38 | Qwen3 VL 30B A3B Reasoning | Alibaba | 83.9 | 1 | 2026-08-23 |
| 39 | InternVL-3.5 (8B) | OpenGVLab | 83.7 | 2 | 2026-06-05 |
| 40 | Qwen3 VL Thinking (8B) | Alibaba | 82.8 | 3 | 2026-08-23 |
| 41 | Mage-VL-4B | Microsoft | 81.8 | 1 | 2026-07-26 |
| 42 | Phi 4 MM 5.6B | Microsoft | 81.7 | 1 | 2026-07-26 |
| 43 | Qwen3 VL 4B (Reasoning) | Alibaba | 81.6 | 2 | 2026-08-23 |
| 44 | GPT-5.2 | OpenAI | 80.7 | 1 | 2026-06-15 |
| 45 | North-Micro-Vision-Instruct | Cohere | 79.2 | 3 | 2026-08-23 |
| 46 | Phi 4 R V 15B | Microsoft | 73.9 | 1 | 2026-07-26 |
| 47 | LFM2.5-VL-1.6B | Liquid AI | 41.4 | 3 | 2026-08-12 |
| 48 | LFM2-VL-1.6B | Liquid AI | 35.1 | 1 | 2026-08-09 |
| 49 | InternVL3.5-1B | OpenGVLab | 33.5 | 1 | 2026-08-09 |
| 50 | Qwen3.5 2B | Alibaba | 0.9 | 2 | 2026-08-12 |
| 51 | Qwen3 VL 2B Instruct | Alibaba | 0.8 | 2 | 2026-08-12 |
| 52 | Ministral 3 3B | Mistral | 0.7 | 2 | 2026-08-12 |
| 53 | Gemma 4 E2B | 0.7 | 2 | 2026-08-12 | |
| 54 | Phi 3.5 Vision Instruct | Microsoft | 0.6 | 2 | 2026-08-12 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.