Optical Character Recognition Benchmarking on OCRBench (test) benchmark leaderboard · SOTA2 Research

Optical Character Recognition Benchmarking on OCRBench (test)

85.4AccuracyQwen2.5-VL-7B + Q-Zoom

Evaluation Results

Method	Links
Qwen2.5-VL-7B + Q-Zoom2026.04		85.4
Qwen3-VL-4B + Q-Zoom2026.04		84.6
Qwen3-VL-4B + SD-RPN2026.04		84.2
Qwen3-VL-4B2026.04		83.1
Qwen2.5-VL-7B + SD-RPN2026.04		82.9
Qwen2.5-VL-7B2026.04		82.8
Qwen2.5-VL-7B + VisionThink2026.04		80.8
Qwen2.5-VL-3B + Q-Zoom2026.04		79.8
Qwen2.5-VL-7B + AdaptVision2026.04		76.9
Qwen2.5-VL-3B + SD-RPN2026.04		76.3
Qwen2.5-VL-3B2026.04		75.9
LLaVA-1.5-13B + SD-RPN2026.04		39.6
LLaVA-1.5-13B + Q-Zoom2026.04		39.4
LLaVA-1.5-7B + Q-Zoom2026.04		38
LLaVA-1.5-7B + SD-RPN2026.04		37.3
LLaVA-1.5-13B + S22026.04		36.4
LLaVA-1.5-13B + ViCrop2026.04		36.1
LLaVA-1.5-13B2026.04		33.7
LLaVA-1.5-7B + ViCrop2026.04		33.2
LLaVA-1.5-7B + S22026.04		32.6
LLaVA-1.5-7B2026.04		31.4