Optical Character Recognition on OCRBench (test)
87.2ScoreQwen3-VL-8B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-VL-8BParameter Scale=8B2026.05 | 87.2 | — | |
| PerceptionLM-8BParameter Scale=8B2026.05 | 84.2 | — | |
| Qwen3-VL-2BParameter Scale=2B2026.05 | 84.1 | — | |
| Qwen3-VL-4BParameter Scale=4B2026.05 | 84.1 | — | |
| InternVL3.5-8BParameter Scale=8B2026.05 | 83.8 | — | |
| InternVL3.5-2BParameter Scale=2B2026.05 | 83.4 | — | |
| InternVL3.5-4BParameter Scale=4B2026.05 | 82 | — | |
| Zamba2-VL-7BParameter Scale=7B2026.05 | 81.6 | — | |
| PerceptionLM-3BParameter Scale=3B2026.05 | 80.1 | — | |
| InternVL3.5-1BParameter Scale=1B2026.05 | 79.2 | — | |
| PerceptionLM-1BParameter Scale=1B2026.05 | 79 | — | |
| InternVL2-2BModel Scale=1B2024.09 | 78.1 | — | |
| Gemini-1.5-ProModel Scale=High-End2024.09 | 75.4 | — | |
| GPT-4oModel Scale=High-End2024.09 | 73.6 | — | |
| Zamba2-VL-2.7BParameter Scale=2.7B2026.05 | 73.6 | — | |
| Zamba2-VL-1.2BParameter Scale=1.2B2026.05 | 71.4 | — | |
| MM1.5-30BModel Scale=30B2024.09 | 65.8 | — | |
| MM1.5-3BModel Scale=3B2024.09 | 65.7 | — | |
| GPT-4VModel Scale=High-End2024.09 | 64.5 | — | |
| MM1.5-3B-MOEModel Scale=3B, Architecture=MoE2024.09 | 63.8 | — | |
| Phi-3-Vision-4BModel Scale=4B2024.09 | 63.7 | — | |
| MM1.5-7BModel Scale=7B2024.09 | 63.5 | — | |
| MM1.5-1B-MOEModel Scale=1B, Architecture=MoE2024.09 | 62.6 | — | |
| MM1-7BModel Scale=7B2024.09 | 62.6 | — | |
| FP16 (Baseline)Bits=FP162026.02 | 62.2 | — | |
| Molmo2-4BParameter Scale=4B2026.05 | 62 | — | |
| Molmo2-8BParameter Scale=8B2026.05 | 61.4 | — | |
| QuEPTBits=W4A82026.02 | 61.2 | — | |
| MBQBits=W32026.02 | 61.1 | — | |
| MM1-30BModel Scale=30B2024.09 | 60.6 | — | |
| QuEPTBits=W32026.02 | 60.6 | — | |
| MM1.5-1BModel Scale=1B2024.09 | 60.5 | — | |
| MiniCPM-V 2.0-3BModel Scale=3B2024.09 | 60.5 | — | |
| AWQBits=W32026.02 | 59.3 | — | |
| MetaGPTunsupervised=true2025.03 | 57.3 | — | |
| MM1-3BModel Scale=3B2024.09 | 57 | — | |
| MM1-1BModel Scale=1B2024.09 | 56.6 | — | |
| CogVLM(base)Model Type=Original2025.03 | 56.5 | — | |
| AdaMMSunsupervised=true2025.03 | 56.3 | — | |
| MBQBits=W4A82026.02 | 52.3 | — | |
| Task Arithmeticunsupervised=false2025.03 | 51.9 | — | |
| DARE-Linearunsupervised=false2025.03 | 50.9 | — | |
| Ties-Mergingunsupervised=false2025.03 | 50.7 | — | |
| QuEPTBits=W4A42026.02 | 49.1 | — | |
| QuEPTBits=W22026.02 | 47.8 | — | |
| DARE-Tiesunsupervised=false2025.03 | 42.9 | — | |
| DeepSeek-VLModel Scale=1B2024.09 | 40.9 | — | |
| mPLUG-Owl2Model Type=Original2025.03 | 34.1 | — | |
| SmoothQBits=W4A82026.02 | 32 | — | |
| COINCIDESelection Budget=100K2026.02 | — | 18.3 | |
| Full-FinetuneSelection Budget=625K (Full)2026.02 | — | 20.3 | |
| Gemma-3-27B-ITJudge=GPT-4o2025.12 | — | 77.6 | |
| InternVL3-8BJudge=GPT-4o2025.12 | — | 88 | |
| LengthSelection Budget=100K2026.02 | — | 13.6 | |
| MiMo-VL-7B-SFT-2508Judge=GPT-4o2025.12 | — | 87.8 | |
| MiMo-VL-Miloco-7BJudge=GPT-4o2025.12 | — | 86.5 | |
| PerplexitySelection Budget=100K2026.02 | — | 18.6 | |
| PRISMSelection Budget=100K2026.02 | — | 18.8 | |
| Qwen2.5-VL-7BJudge=GPT-4o2025.12 | — | 89.7 | |
| RandomSelection Budget=100K2026.02 | — | 17.9 | |
| RDS+Selection Budget=100K2026.02 | — | 18.6 | |
| ScalSelectSelection Budget=50K2026.02 | — | 18.1 | |
| ScalSelectSelection Budget=100K2026.02 | — | 19.4 | |
| ScalSelectSelection Budget=200K2026.02 | — | 19.5 | |
| ScalSelectSelection Budget=300K2026.02 | — | 19.3 | |
| ScalSelectSelection Budget=400K2026.02 | — | 19.8 |