Share your thoughts, 1 month free Claude Pro on usSee more

Physical Reasoning on PhysBench

97.1S1 Score

Human

Updated 1mo ago

Evaluation Results

Method	Links
Human 2026.04		97.1	95.67	94.91	95.68	95.87
PhysNote 2026.04		62.45	72.1	42.33	49.85	56.68
PhysAgent 2026.04		58.2	65.4	38.1	45.2	51.72
Gemini-1.5-flash 2026.04		57.41	52.24	34.32	40.93	46.07
Gemini-1.5-pro 2026.04		57.26	63.61	36.52	41.56	49.11
GPT-4o 2026.04		56.91	64.8	30.15	46.99	49.49
InternVL-Chat-1.5 2026.04		53.08	70.14	37.01	44.78	47.51
LLaVA-interleave-dpo 2026.04		47.97	42.67	33.73	38.78	40.83
Claude-3.5-sonnet 2026.04		46.46	41.11	27.89	37.6	38.05
Mantis-siglip-llama3 2026.04		42.47	32.78	36.83	37.51	37.64
LLaVA-1.5-13B 2026.04		41.31	42.5	34.4	44.38	40.45
LLaVA-1.6-vicuna-7B 2026.04		40.26	59.72	38.6	42.65	42.28
PLLaVA-13B 2026.04		39.91	38.33	31.52	40.76	37.7