Mental Health Safety Evaluation on MHSafeEval (v1.0) benchmark leaderboard · SOTA2 Research

Mental Health Safety Evaluation on MHSafeEval (v1.0)

37.5ASR (Gaslighting)MiniMax m2.5

Evaluation Results

Method	Links
MiniMax m2.52026.04		37.5	50	28.6	22.2	77.8	87.5	58.8	52.9	15.7	61.4
Llama 3.12026.04		54.3	50	21.6	27.8	81.4	96.6	75	58.9	57	97.3
GPT-3.52026.04		60	50.9	39.3	27.6	70.8	100	72.7	60.3	38.4	99.5
MiMo2026.04		71.9	54.3	27	57.6	90.6	95	71.2	64.9	38.9	98.9
Gemini 2.52026.04		79.6	68.8	47.5	63.5	90.7	100	59.6	70.8	19.5	99.2
DeepSeek v3.22026.04		80.9	77.8	59.5	59.5	93.8	94.3	73.8	76.2	20	99.5
Haiku 4.52026.04		88.2	87.8	60.7	46.9	94.2	86.2	87.1	78.9	83.2	99.5
Gemma 42026.04		90	94.7	66.7	70.9	100	100	84	87.3	15.7	97.8
MiniMax m2.52026.04		93.3	86	81.5	83.8	92.9	98	95.5	91.4	3	81.1
Gemini 2.52026.04		93.8	98.3	92.6	98.3	98.2	100	95.9	97	3.8	97.3
DeepSeek v3.22026.04		93.8	97	94.4	95.7	100	100	96.7	97	12.4	99.7
Llama 3.12026.04		94.5	87.5	79.3	83.8	89.7	98.8	98.3	92.2	55.7	94.1
MiMo2026.04		95.2	96	93.9	92.8	95.6	96.4	90.9	94.3	34.3	99.7
Haiku 4.52026.04		98.3	100	96.9	91.7	98.2	95.9	97.1	97	85.9	98.6
GPT-3.52026.04		100	100	66.7	75	100	100	100	94.3	7.1	100
Gemma 42026.04		100	97.7	100	100	100	100	100	99.7	7	95.9