Prompt Injection Defense on Three Suites Aggregated 1.0 (test) benchmark leaderboard · SOTA2 Research

Prompt Injection Defense on Three Suites Aggregated 1.0 (test)

89.5Clean UtilityDeepSeek-v2-Flash

Evaluation Results

Method	Links
DeepSeek-v2-Flash2026.06		89.5	22.6	77.4
DeepSeek-v2-Flash2026.06		89.5	16.5	77.9
GPT-3.5-turbo2026.06		86	2.8	67.6
GPT-3.5-turbo2026.06		86	5.4	66.6
GPT-3.5-turbo2026.06		77.2	6.9	60.2
GPT-3.5-turbo2026.06		77.2	8.7	62.2
Claude-3-Haiku2026.06		70.2	0.3	61.4
Claude-3-Haiku2026.06		70.2	1.8	64
GPT-4o-mini2026.06		68.4	39.3	46.8
GPT-4o-mini2026.06		68.4	40.4	42.9
GPT-4o-mini2026.06		66.7	58.6	42.4
GPT-4o-mini2026.06		66.7	52.4	41.4
GPT-4o-mini2026.06		64.9	12.6	53.2
GPT-4o-mini2026.06		64.9	33.4	45.8
Gemini-1.5-Flash2026.06		63.2	47.8	40.1
Gemini-1.5-Flash2026.06		63.2	32.4	44.5
GPT-4o-mini2026.06		63.2	54.2	37.3
GPT-4o-mini2026.06		63.2	49.4	42.4
GPT-4o-mini2026.06		63.2	8.2	47.6
GPT-4o-mini2026.06		63.2	7.7	45.8
GPT-4o-mini2026.06		59.6	42.7	46.5
GPT-4o-mini2026.06		59.6	36.3	48.6
GPT-4o-mini2026.06		59.6	51.7	43.4
GPT-4o-mini2026.06		59.6	46.8	42.4
GPT-4o-mini2026.06		49.1	2.6	44.7
GPT-4o-mini2026.06		49.1	6.4	42.4
GPT-4o-mini2026.06		47.4	7.2	30.3
GPT-4o-mini2026.06		47.4	15.4	45
Claude-3-Haiku2026.06		45.6	0.3	47
Claude-3-Haiku2026.06		45.6	1.5	44.7
GPT-4o-mini2026.06		43.9	0	30.6
GPT-4o-mini2026.06		43.9	28	34.4