Pedagogical Dialogue Evaluation on SHAPE (test) benchmark leaderboard · SOTA2 Research

Pedagogical Dialogue Evaluation on SHAPE (test)

93.66Safety ScoreClaude Opus 4.5

Evaluation Results

Method	Links
Claude Opus 4.52026.04		93.66	100	74.44
Claude Haiku 4.52026.04		93.66	100	76.69
Gemini 2.5 Pro2026.04		93.66	100	72.18
Gemini 2.5 Flash2026.04		93.66	100	82.71
Gemini 2.5 Flash-Lite2026.04		93.66	100	78.2
Claude Haiku 4.52026.04		92.96	100	74.24
Qwen3-80B2026.04		92.96	100	74.24
Claude Opus 4.52026.04		92.25	100	68.7
Gemini 2.5 Pro2026.04		92.25	100	71.76
Gemini 2.5 Flash2026.04		92.25	100	75.57
Qwen3-80B2026.04		92.25	100	70.99
Claude Opus 4.52026.04		91.55	100	73.08
Gemini 2.5 Flash-Lite2026.04		91.55	100	83.85
Gemini 2.5 Flash-Lite2026.04		90.85	100	83.72
Qwen3-80B2026.04		90.85	100	65.89
GPT-5 mini2026.04		90.14	100	93.75
Claude Haiku 4.52026.04		89.44	100	66.93
GPT-52026.04		89.44	100	94.49
GPT-5 mini2026.04		88.46	100	89.13
GPT-52026.04		86.62	100	94.31
GPT-5 mini2026.04		86.62	100	93.5
Qwen3-32B2026.04		86.62	100	78.86
Qwen3-8B2026.04		86.62	100	74.8
GPT-52026.04		85.92	100	95.08
GPT-5 nano2026.04		79.58	100	86.73
Gemini 2.5 Pro2026.04		77.46	98.28	75.45
GPT-5 nano2026.04		77.46	100	90.91
GPT-5 nano2026.04		73.94	100	84.76
Gemini 2.5 Flash2026.04		39.44	98.28	71.43
Qwen3-32B2026.04		15.49	100	63.64
Qwen3-32B2026.04		7.04	100	50
Qwen3-8B2026.04		3.52	100	40
Qwen3-8B2026.04		1.41	100	0