GPT
Benchmarks
TaskDatasetSOTA resultTrendResults
MiniGPT-4
7.32Attack Success Rate (ASR)
36
GPT-4o
0.99ASR
19
GPT-4o
0.6ASR
14
GPT 5.1
96.5ASR
13
GPT 5.4
0ASR
12
GPT Small (val)
27.95Validation Perplexity
12
GPT 5.2 (External Holdout)
88.3AUROC
12
GPT image 2
0ASR
11
GPT
23.8Tuning Time (s)
9
GPT-5
18.67Attack Success Rate
9
GPT 4.1
4.33Attack Success Rate (ASR)
9
GPT closed-source standard MLLMs 5.4
3.8ASR
8
GPT closed-source standard MLLMs 5.2
3.3ASR
8
GPT-4o 5.2 (test)
46.9Attack Success Rate (ASR)
8
GPT Pre-training (val)
19.98Validation Perplexity
8
GPT Paraphrased 4.1
0.398ROC AUC (1% FPR)
8
GPT nano (val)
3.25Validation Loss
5
GPT 4.1 8 July 2025 release
99.8ASR
5
GPT-G
76.8Semantic Objective
4
GPT-3.5 (test)
99.14Accuracy
4
GPT 4.1
100Detection Accuracy (No Attack)
3
GPT generated summaries 5.1
86.5BERTScore-F1
3
MiniGPT-4
48.1BLEU-1
3
GPT Clean Holdout 5.2 (test)
0.886AUROC
1