Method MangaOCR
Hmean (%)
MangaVQA
LLM (/10.0)
Proprietary Models
GPT-4o 0.0 6.00
Gemini 2.5 Flash 0.0 7.26
Claude Sonnet 4.5 0.0 5.84
Open-source Models
Phi-4-Multimodal-5.6B 0.0 3.39
Pangea-7B 0.0 3.23
LLaVA-OV-1.5-8B 0.0 3.46
Sarashina2-Vision-8B 0.0 4.45
Gemma-3-12B 0.0 4.13
Heron-NVILA-Lite-15B 0.0 3.76
Qwen2.5-VL-7B 0.9 5.65
Ours
MangaLMM (Ours) 71.5 6.68