Initial Evaluations of MMCTAgent Image Dataset Claude 3 Opus* Claude 3 Sonnet* Claude 3 Haiku* GPT-4V* Gemini 1.0 Ultra* Gemini 1.5 Pro* Gemini 1.0 Pro* MMCT w/o Critic MMCT w Critic MMMU 59.40 53.10 50.20 56.80 59.40 58.50 47.90 59.34 63.57 MathVista 50.50 47.90 46.40 49.90 53.00 52.10 45.20 53.30 56.50 MMVET 51.70 51.30 - 60.20 - 64.20 - 70.51 74.24 MMBench 63.30 67.80 60.70 77.00 - 73.60 - 80.21 84.20 For EgoSchema video Dataset: Method Acc. GPT-4V [2] 63.5 Gemini 1.0 Pro [32] 61.5 LLoVi [46] 57.6 MC-ViT-L [6] 62.6 Video-LLaVa [14] 36.8 ViperGPT [31] 15.8 VideoAgent [34] 60.2 VideoAgent-M [9] 62.8 MMCT w/o Critic 68.8 MMCT w Critic 71.2 Current Evaluations Image Datasets GPT-4V MMCT with GPT-4V GPT-4o MMCT with GPT-4o GPT-5 MMCT with GPT-5 MM-Vet 60.20 74.24 77.98 79.36 80.51 81.65 MMMU 56.80 63.57 69.10 73.00 84.20 85.44 Video Datasets GPT-4o MMCT with GPT-4o VideoMME 72.10 76.70