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
100.0%
#2
Claude Sonnet 4.5 (Thinking)
100.0%
#3
Claude Sonnet 4.5
100.0%
#4
Claude Haiku 4.5 (Thinking)
100.0%
#5
GPT-4.1
100.0%
#6
Claude Haiku 4.5
100.0%
#7
Grok 4
100.0%
#8
Grok 3
100.0%
#9
GPT OSS 120B
100.0%
#10
Grok 3 (Thinking)
100.0%
#11
Grok 4 (Thinking)
100.0%
#12
OpenAI o4 Mini (High Effort)
100.0%
#13
OpenAI o4 Mini (Medium Effort)
100.0%
#14
OpenAI o3 (High Effort)
97.2%
#15
Claude 3.5 Haiku
97.2%
#16
Claude 3 Sonnet
97.2%
#17
OpenAI o3 (Medium Effort)
97.2%
#18
Claude 3 Opus
95.8%
#19
Granite 3.0
95.0%
#20
Claude 4.0 Opus
92.0%
#21
Claude 4.0 Opus (Thinking)
92.0%
#22
Kimi K2
91.7%
#23
Claude 4.0 Sonnet
90.7%
#24
Claude 4.0 Sonnet (Thinking)
90.7%
#25
GPT-5 (Thinking)
90.0%
#26
GPT-5
90.0%
#27
OpenAI o1
90.0%
#28
OpenAI o3 Mini (Medium Effort)
88.9%
#29
OpenAI o3 Mini (High Effort)
88.9%
#30
Gemini 2.5 Pro (Thinking)
87.7%
#31
Gemini 2.5 Pro
87.7%
#32
GPT-3.5 Turbo
84.6%
#33
Gemini 2.0 Pro
83.3%
#34
GPT-4.5
83.3%
#35
GPT-4o Mini
83.3%
#36
Gemini 2.0 Flash (Thinking)
80.6%
#37
Gemini 2.0 Flash Thinking (Jan 2025)
80.6%
#38
Gemini 2.0 Flash
80.6%
#39
Gemini 3.1 Pro Preview
77.8%
#40
Claude 3.7 Sonnet
76.4%
#41
Claude 3.7 Sonnet (Thinking)
76.4%
#42
Gemini 2.5 Flash (Thinking)
75.0%
#43
Gemini 2.5 Flash
75.0%
#44
Llama 4 Maverick 17B
72.2%
#45
GPT-4o
69.4%
#46
Llama 3.1 405B
66.7%
#47
Gemini 1.5 Pro
63.9%
#48
Qwen 3
52.8%
#49
Llama 3.3 70B
52.8%
#50
Qwen 3 (Thinking)
52.8%
#51
DeepSeek V3
50.0%
#52
DeepSeek R1
40.3%
#53
Grok 3 Mini
40.0%
#54
Mistral Large 2
36.1%
#55
OpenAI o1 Mini
33.3%
#56
Cohere Command A
19.4%
#57
Cohere Command R+
19.4%