Insight-V++: Towards Advanced Long-Chain Visual Reasoning with Multimodal Large Language Models

Benchmark Model Rank Results
chart-question-answering-on-chartqaInsight-V++–1:1 Accuracy: 86.1
mathematical-reasoning-on-mathverseInsight-V++–Accuracy: 62.4
multimodal-reasoning-on-mathvisionInsight-V++–Accuracy: 48.6
multimodal-reasoning-on-mathvistaInsight-V++–Accuracy: 77.6
multimodal-reasoning-on-mmmuInsight-V++–Accuracy: 64.8
multimodal-reasoning-on-mmmu-proInsight-V++–Accuracy: 45.6
video-question-answering-on-mmvuInsight-V++–Accuracy: 65.0
video-question-answering-on-videoholmesInsight-V++–Accuracy: 40.2
video-question-answering-on-videommmuInsight-V++ (Self-Evolving, Qwen2.5-VL, 7B)–Accuracy: 57.6
zero-shot-video-question-answer-on-video-mme-w-o-subsInsight-V++ (Self-Evolving, Qwen2.5-VL, 7B)–Accuracy (%): 67.8