. 2026 Jul 7;26:1889. doi: 10.1186/s12903-026-09174-w

Table 1.

Comparison of correct response rates within and between models

	ChatGPT 5.2			Gemini 3			DeepSeek V3.2			Test statistics	p
	Value (%) p	CI 95%-Wald binominal	Standard error	Value (%) p	CI 95%-Wald binominal	Standard error	Value (%) p	CI 95%-Wald binominal	Standard error	Test statistics	p
Group A	108 (90) ^aA	84,632: 95,368	0,027	106 (88,3) ^abA	82,590: 94,077	0,029	93 (77,5) ^b	70,029: 84,971	0,038	8,253	0,016^x
Group B	107 (89,2) ^A	83,606: 94,728	0,028	98 (81,7) ^A	74,744: 88,590	0,035	95 (79,2)	71,900: 86,433	0,037	4,833	0,109^x
Group C	104 (86,7) ^A	80,585: 92,749	0,031	107 (89,2) ^A	83,606: 94,728	0,028	106 (88,3)	82,590: 94,077	0,029	0,387	0,880^x
Group D	119 (99,2) ^aB	97,540: 100,000	0,008	120 (100) ^aB	00,000: 100,000	0,000	103 (85,8) ^b	79,594: 92,072	0,032	29,225	< 0,001^x
Group E	94 (78,3) ^aA	70,962: 85,704	0,038	108 (90) ^bA	84,632: 95,368	0,027	103 (85,8) ^ab	79,594: 92,072	0,032	6,299	0,048^x
Group F	108 (90) ^aA	84,632: 95,368	0,027	108 (90) ^aA	84,632: 95,368	0,027	91 (75,8) ^b	68,174: 83,493	0,039	11,845	0,001^x
Total	640 (88,9) ^a	86,593: 91,184	0,012	647 (89,9) ^a	87,656: 92,066	0,011	591 (82,1) ^b	79,282: 84,884	0,014	22,783	< 0,001^y
Test statistics	30,936			29,824			11,023
p	< 0,001^x			< 0,001^x			0,058^x

^x Fisher’s Exact Test with Monte Carlo correction

^y Pearson’s chi-square test; frequency (percentage)

^a–b No significant difference between language models sharing the same letter

^A–B No significant difference between groups sharing the same letter. Bold values indicate statistically significant differences (p < 0.05).