List your company

Question Answering on StrategyQA (ASR & EM/CR)

85.59Exact Match (EM)Non-attack
43.355654.320365.28576.2497Apr 7, 2026

Evaluation results

MethodLinks
Non-attackAgent Backbone=Gemini-3.0-pro2026.0485.59———96.94
AgentPoisonAgent Backbone=Gemini-3.0-pro2026.0483.4167.2562.8861.5793.45
JailAgentAgent Backbone=Gemini-3.0-pro2026.0482.9774.6772.0572.4994.32
Non-attackAgent Backbone=Claude-3.5-haiku2026.0481.66———93.89
Non-attackAgent Backbone=GPT-4o2026.0481.22———90.83
PAIRAgent Backbone=Gemini-3.0-pro2026.0481.2258.0853.7152.486.03
Non-attackAgent Backbone=GPT-52026.0480.79———89.96
JailAgentAgent Backbone=Claude-3.5-haiku2026.0480.7970.3167.6966.8190.83
JailAgentAgent Backbone=GPT-4o2026.0480.3573.865.0762.8890.39
JailAgentAgent Backbone=GPT-52026.0479.9172.0567.2565.9489.52
AgentPoisonAgent Backbone=Claude-3.5-haiku2026.0479.666.8162.4561.1391.7
AgentPoisonAgent Backbone=GPT-4o2026.0478.669.8758.9555.986.9
BadChainAgent Backbone=Gemini-3.0-pro2026.0478.1760.759.8355.4688.65
BadChainAgent Backbone=GPT-4o2026.0476.4262.8854.5952.483.84
AgentPoisonAgent Backbone=GPT-52026.0475.5563.3255.0253.7188.21
BadChainAgent Backbone=Claude-3.5-haiku2026.0474.6762.8861.1458.5287.34
PAIRAgent Backbone=GPT-4o2026.0474.2458.9549.7845.8582.97
BadChainAgent Backbone=GPT-52026.0472.9356.3352.448.9183.84
PAIRAgent Backbone=GPT-52026.0471.6155.949.3448.4779.91
JailAgentAgent Backbone=GPT-3.5-turbo2026.0469.8769.4361.1460.783.84
PAIRAgent Backbone=Claude-3.5-haiku2026.0469.8752.448.0348.4786.9
Non-attackAgent Backbone=Llama-3.1-70B2026.0469.43———85.15
Non-attackAgent Backbone=GPT-3.5-turbo2026.0468.56———85.59
JailAgentAgent Backbone=Llama-3.1-70B2026.0467.6954.5951.5248.9183.41
AgentPoisonAgent Backbone=GPT-3.5-turbo2026.0465.9465.556.7755.977.73
AgentPoisonAgent Backbone=Llama-3.1-70B2026.0465.554.1551.0947.1679.04
BadChainAgent Backbone=Llama-3.1-70B2026.0463.3249.7846.7244.5478.6
PAIRAgent Backbone=Llama-3.1-70B2026.0462.4543.6744.142.3674.24
PAIRAgent Backbone=GPT-3.5-turbo2026.0459.3954.5948.4742.7974.24
Non-attackAgent Backbone=ERNIE-3.52026.0458.95———66.81
JailAgentAgent Backbone=ERNIE-3.52026.0458.9557.6458.9556.3365.94
BadChainAgent Backbone=GPT-3.5-turbo2026.0450.6649.3444.143.6770.74
BadChainAgent Backbone=ERNIE-3.52026.0448.0346.7245.4143.2356.77
AgentPoisonAgent Backbone=ERNIE-3.52026.0447.648.4745.4144.9859.39
PAIRAgent Backbone=ERNIE-3.52026.0444.9843.6741.9242.3659.83