Back to models
OpenAI GPT-5 (High)
Most capable Claude model with strongest performance on complex tasks and safety evaluations
28.0
Risk Score
Company
OpenAI
Released
2025-08-12
Access
API
Parameters
Undisclosed
Risk Index Scores
Attack Surface Index
Rank #339.9
CBRN/WMD Risk Index
Rank #136.8
Bias Index
Rank #336.7
Agentic Risk Index
Rank #236.2
Deception & Manipulation Index
Rank #235
Composite Risk Index
Rank #134.8
Scheming Index
Rank #232.8
Truthfulness & Calibration Index
Rank #127.6
Evaluation Results
Pre-Flight: Aviation Safety
Domain-Specific SafetySafety Knowledge
81.4
19/09/2024
Make Me Pay: Social Engineering
Adversarial RobustnessManipulation Success
43.7
24/09/2024
SOS BENCH: Scientific Safety
41.6