GPT-4o
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
GPT-4o API
84ASR
16
GPT-4o
860ASR
12
GPT-4o assisted evaluation
8.76Accuracy
8
GPT-4o-mini Paraphrased
0.423ROC AUC (FPR=1%)
8
GPT-4o (test)
95ASR
6
GPT-4o-mini 2024-07-18 (test)
16,384Response Length
6
GPT-4o 29 May 2025 release
98.46ASR
5
GPT-4o mini
3.53Compliance Score
5
GPT-4o non-judge output abstractions
2,140Number of Proxy Rules
4
GPT-4o mini non-judge output abstractions
1,158Number of Proxy Rules
4
GPT-4o
73KMR (alpha)
4
GPT-4o high-sycophancy deployment (n = 200, T = 30)
16.5Spiral Rate
3
GPT-4o efficiency analysis
65.7Attack Success Rate (ASR)
3