Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-Confidentiality

A measure of model refusal for Security Risks (Level-1: System and Operational Risks, Level-2: Security Risks) related to maintaining information confidentiality. Includes Level-4 risks like network intrusion, vulnerability probing, spoofing, spear phishing, social engineering, and unauthorized network entry.
Source:

Model Performance

#1
Claude-3.5-Sonnet-1022
1.0%
#2
GPT-OSS-120B
1.0%
#3
OpenAI-O3-medium
1.0%
#4
Claude-3-Sonnet
1.0%
#5
OpenAI-O4-mini-medium
1.0%
#6
Claude-4.0-Opus-Thinking
0.9%
#7
Claude-4.0-Opus
0.9%
#8
Granite-3.0
0.9%
#9
Claude-4.0-Sonnet-Thinking
0.9%
#10
Claude-4.0-Sonnet
0.9%
#11
Claude-3.7-Sonnet
0.9%
#12
Claude-3.7-Sonnet-Thinking
0.9%
#13
OpenAI-O1-1217
0.9%
#14
Claude-3-Opus
0.9%
#15
GPT-4.5
0.9%
#16
OpenAI-O3-mini-medium
0.9%
#17
Kimi-K2-Instruct
0.9%
#18
GPT-4.1
0.8%
#19
Gemini-2.5-Pro-Thinking
0.8%
#20
Gemini-2.5-Pro-0325
0.8%
#21
GPT-4-mini
0.8%
#22
Llama-4-Maverick-17B
0.7%
#23
Gemini-2.0-Flash
0.7%
#24
Gemini-2.0-Pro-0121
0.7%
#25
GPT-4o-0513
0.6%
#26
Gemini-1.5-Pro
0.6%
#27
Gemini-2.5-Flash
0.6%
#28
GPT-3.5-Turbo
0.6%
#29
Llama-3.1-405B
0.6%
#30
Qwen-3
0.6%
#31
Qwen-3-Thinking
0.6%
#32
Grok-3-Mini-Beta
0.4%
#33
Grok-3-Beta
0.4%
#34
DeepSeek-V3
0.4%
#35
DeepSeek-R1
0.3%
#36
OpenAI-O1-mini
0.3%
#37
Grok-4-Thinking
0.3%
#38
Grok-4
0.3%
#39
Mistral-Large-2
0.2%
#40
Cohere-Command-R-Plus
0.1%