List your company

Trace-level safety monitoring on ImpossibleBench

95.7ROC-AUCMeerkat
48.17260.51172.8585.189Apr 13, 2026

Evaluation results

MethodLinks
MeerkatTPC=100, Model=GPT-5.4m2026.0495.7
MeerkatTPC=100, Model=Qwen3.52026.0495.1
BufferTPC=100, Model=GPT-5.4m2026.0494.9
BufferTPC=100, Model=Qwen3.52026.0494.8
MeerkatTPC=50, Model=GPT-5.4m2026.0494.4
MonitorTPC=100, Model=GPT-5.4m2026.0494.3
BayesianTPC=100, Model=Qwen3.52026.0493.5
BayesianTPC=100, Model=GPT-5.4m2026.0493.5
MonitorTPC=50, Model=GPT-5.4m2026.0493.3
MeerkatTPC=50, Model=Qwen3.52026.0492.8
MonitorTPC=10, Model=GPT-5.4m2026.0492.4
MeerkatTPC=25, Model=GPT-5.4m2026.0492.4
BufferTPC=50, Model=GPT-5.4m2026.0491.5
BufferTPC=50, Model=Qwen3.52026.0491.4
MonitorTPC=25, Model=GPT-5.4m2026.0490
MeerkatTPC=10, Model=Qwen3.52026.0489.8
MeerkatTPC=10, Model=GPT-5.4m2026.0489.8
MonitorTPC=100, Model=Qwen3.52026.0489.3
BayesianTPC=50, Model=Qwen3.52026.0488
BayesianTPC=50, Model=GPT-5.4m2026.0488
BayesianTPC=10, Model=Qwen3.52026.0487.8
BayesianTPC=10, Model=GPT-5.4m2026.0487.8
MeerkatTPC=25, Model=Qwen3.52026.0487.6
BufferTPC=10, Model=GPT-5.4m2026.0487.3
BufferTPC=10, Model=Qwen3.52026.0487.1
MonitorTPC=50, Model=Qwen3.52026.0487
BufferTPC=25, Model=Qwen3.52026.0485.5
BufferTPC=25, Model=GPT-5.4m2026.0485.5
BayesianTPC=25, Model=Qwen3.52026.0484.8
BayesianTPC=25, Model=GPT-5.4m2026.0484.8
MonitorTPC=10, Model=Qwen3.52026.0480
Naive AgentTPC=50, Model=Qwen3.52026.0478.9
MonitorTPC=25, Model=Qwen3.52026.0478
Naive AgentTPC=100, Model=GPT-5.4m2026.0474.2
Naive AgentTPC=10, Model=Qwen3.52026.0473.8
Naive AgentTPC=25, Model=Qwen3.52026.0471.8
Naive AgentTPC=100, Model=Qwen3.52026.0470
Naive AgentTPC=50, Model=GPT-5.4m2026.0461.9
Naive AgentTPC=10, Model=GPT-5.4m2026.0450
Naive AgentTPC=25, Model=GPT-5.4m2026.0450