List your company

Question Answering on MMLU (ASR-R, ASR-L, ASR-H, EM, CR)

72.28EMNon-attack
28.121639.585851.0562.5142Apr 7, 2026

Evaluation results

MethodLinks
Non-attackAgent Backbone=GPT-52026.0472.28———85.96
JailAgentAgent Backbone=GPT-52026.0471.9370.1866.8465.7986.32
JailAgentAgent Backbone=GPT-4o2026.0469.8265.6164.2163.1685.61
Non-attackAgent Backbone=GPT-4o2026.0469.65———85.09
AgentPoisonAgent Backbone=GPT-4o2026.0465.6162.8156.6756.3284.56
BadChainAgent Backbone=GPT-4o2026.0464.9161.460.3558.4282.63
AgentPoisonAgent Backbone=Gemini-3.0-pro2026.0463.1660.3557.5457.5487.02
JailAgentAgent Backbone=Gemini-3.0-pro2026.0462.9862.8159.6557.7291.23
JailAgentAgent Backbone=GPT-3.5-turbo2026.0462.4660.5358.4257.5482.28
Non-attackAgent Backbone=GPT-3.5-turbo2026.0462.28———82.81
Non-attackAgent Backbone=Gemini-3.0-pro2026.0462.28———90.35
AgentPoisonAgent Backbone=GPT-52026.0461.5857.5454.0454.5681.58
BadChainAgent Backbone=GPT-52026.0461.454.3952.9853.1680.7
AgentPoisonAgent Backbone=GPT-3.5-turbo2026.0458.9556.4952.6352.8177.89
BadChainAgent Backbone=Gemini-3.0-pro2026.0458.9457.7256.4956.8487.54
JailAgentAgent Backbone=Claude-3.5-haiku2026.0457.5452.2850.5350.784.21
AgentPoisonAgent Backbone=Claude-3.5-haiku2026.0456.4146.3242.2841.7581.4
Non-attackAgent Backbone=Claude-3.5-haiku2026.0456.32———84.56
BadChainAgent Backbone=GPT-3.5-turbo2026.0452.6350.3548.648.0777.19
PAIRAgent Backbone=GPT-4o2026.0452.4645.4442.2842.1168.07
PAIRAgent Backbone=GPT-52026.0450.3550.744.5644.9175.09
PAIRAgent Backbone=Gemini-3.0-pro2026.0450.3544.6537.3737.0277.02
PAIRAgent Backbone=GPT-3.5-turbo2026.0449.342.8140.738.9567.89
BadChainAgent Backbone=Claude-3.5-haiku2026.0446.3244.9142.4640.776.32
JailAgentAgent Backbone=Llama-3.1-70B2026.0442.9840.8838.9537.1971.93
Non-attackAgent Backbone=Llama-3.1-70B2026.0442.81———71.93
AgentPoisonAgent Backbone=Llama-3.1-70B2026.0440.8838.9535.2635.4466.32
BadChainAgent Backbone=Llama-3.1-70B2026.0440.5338.2432.8132.9864.04
Non-attackAgent Backbone=ERNIE-3.52026.0438.77———52.28
PAIRAgent Backbone=Claude-3.5-haiku2026.0438.2536.6732.9832.4671.58
JailAgentAgent Backbone=ERNIE-3.52026.0437.0234.9132.4631.5851.75
PAIRAgent Backbone=Llama-3.1-70B2026.0432.8131.9329.8229.8250.7
BadChainAgent Backbone=ERNIE-3.52026.0432.2831.0529.1228.4249.65
AgentPoisonAgent Backbone=ERNIE-3.52026.0430.5330.729.4728.7748.07
PAIRAgent Backbone=ERNIE-3.52026.0429.8225.6122.1121.9342.46