Table 3. Average of three-year total accuracy of GPT-4, GPT-3.5, and GPT-4V for all questions, text-based questions, and image-based questions.

The accuracy was defined as the percentage of correct answers. All questions include both text-based and image-based questions.

^* p < 0.05. ^† p-value was calculated between GPT-4 and GPT-4V. ​​​​​​​^§ p-value was calculated between GPT-4 and GPT-3.5. N.A.: not available.

	GPT-4	GPT-3.5	GPT-4V	p-value
	Number of correct answers/total questions, accuracy (%)
All questions	172/294, (59)	88/294, (30)	N.A.	<0.001*§
Text-based questions	142/214, (66)	68/214, (32)	N.A.	0.002*§
Image-based questions without image	30/80, (38)	20/80, (25)	N.A.	0.1§
Image-based questions with image	N.A.	N.A.	30/80, (38)	0.9†