Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-Availability

A measure of model refusal for Security Risks (Level-1: System and Operational Risks, Level-2: Security Risks) related to system availability. Includes Level-4 risks like system/website impairment and network disruption.
Source:

Model Performance

#1
Claude-3.5-Sonnet-1022
100.0%
#2
GPT-OSS-120B
100.0%
#3
OpenAI-O4-mini-medium
100.0%
#4
Claude-3-Sonnet
97.2%
#5
OpenAI-O3-medium
97.2%
#6
Claude-3-Opus
95.8%
#7
Claude-4.0-Sonnet-Thinking
94.4%
#8
Claude-4.0-Sonnet
94.4%
#9
OpenAI-O1-1217
91.7%
#10
Kimi-K2-Instruct
91.7%
#11
Claude-4.0-Opus
90.3%
#12
Claude-4.0-Opus-Thinking
90.3%
#13
Granite-3.0
90.3%
#14
OpenAI-O3-mini-medium
88.9%
#15
GPT-4.5
83.3%
#16
GPT-4-mini
83.3%
#17
Gemini-2.0-Flash
80.6%
#18
GPT-4.1
77.8%
#19
Gemini-2.5-Pro-Thinking
77.8%
#20
Gemini-2.5-Pro-0325
77.8%
#21
Claude-3.7-Sonnet-Thinking
76.4%
#22
Claude-3.7-Sonnet
76.4%
#23
Gemini-2.5-Flash
75.0%
#24
Gemini-2.0-Pro-0121
75.0%
#25
GPT-3.5-Turbo
75.0%
#26
Llama-4-Maverick-17B
72.2%
#27
GPT-4o-0513
69.4%
#28
Llama-3.1-405B
66.7%
#29
Gemini-1.5-Pro
63.9%
#30
Grok-3-Beta
59.7%
#31
Qwen-3-Thinking
52.8%
#32
Qwen-3
52.8%
#33
DeepSeek-V3
50.0%
#34
Grok-3-Mini-Beta
41.7%
#35
DeepSeek-R1
40.3%
#36
Grok-4-Thinking
36.1%
#37
Mistral-Large-2
36.1%
#38
Grok-4
36.1%
#39
OpenAI-O1-mini
33.3%
#40
Cohere-Command-R-Plus
19.4%