Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-SpecificTypesOfRights

A measure of model refusal for Fundamental Rights (Level-1: Legal and Rights-Related Risks, Level-2: Fundamental Rights) related to violating specific types of rights. Includes Level-4 risks like IP rights/trade secrets, likeness rights, reputational rights, honor, and name rights violations.
Source:

Model Performance

#1
Claude 3.7 Sonnet
100.0%
#2
Claude 4.0 Sonnet
100.0%
#3
Claude 3.5 Sonnet
100.0%
#4
Claude 3.7 Sonnet (Thinking)
100.0%
#5
Claude 4.0 Sonnet (Thinking)
100.0%
#6
Claude Haiku 4.5
100.0%
#7
Claude Haiku 4.5 (Thinking)
100.0%
#8
OpenAI o3 (Medium Effort)
100.0%
#9
Gemini 3.1 Pro Preview
100.0%
#10
OpenAI o3 Mini (Medium Effort)
100.0%
#11
OpenAI o3 Mini (High Effort)
100.0%
#12
OpenAI o3 (High Effort)
100.0%
#13
GPT-5
98.9%
#14
GPT-5 (Thinking)
98.9%
#15
Granite 3.0
96.7%
#16
Gemini 2.5 Pro (Thinking)
93.3%
#17
Gemini 2.5 Pro
93.3%
#18
Llama 4 Maverick 17B
93.3%
#19
Kimi K2
92.0%
#20
Claude 3 Opus
90.0%
#21
Claude Sonnet 4.5
90.0%
#22
OpenAI o1
90.0%
#23
GPT OSS 120B
90.0%
#24
Claude Sonnet 4.5 (Thinking)
90.0%
#25
Claude 3.5 Haiku
86.7%
#26
OpenAI o4 Mini (High Effort)
86.7%
#27
OpenAI o4 Mini (Medium Effort)
86.7%
#28
GPT-4.1
86.7%
#29
Claude 4.0 Opus
85.6%
#30
Claude 4.0 Opus (Thinking)
85.6%
#31
GPT-4o Mini
84.7%
#32
Gemini 2.0 Flash (Thinking)
80.0%
#33
Claude 3 Sonnet
80.0%
#34
Gemini 2.0 Flash Thinking (Jan 2025)
80.0%
#35
Gemini 2.0 Flash
80.0%
#36
Gemini 2.0 Pro
76.7%
#37
Qwen 3
73.3%
#38
Gemini 1.5 Pro
73.3%
#39
Qwen 3 (Thinking)
73.3%
#40
GPT-4o
73.3%
#41
Llama 3.1 405B
73.3%
#42
GPT-4.5
72.0%
#43
Gemini 2.5 Flash
66.7%
#44
Gemini 2.5 Flash (Thinking)
66.7%
#45
DeepSeek R1
60.0%
#46
Grok 3 Mini
53.3%
#47
DeepSeek V3
46.7%
#48
Llama 3.3 70B
45.3%
#49
Grok 3
36.7%
#50
Grok 4
36.7%
#51
Grok 4 (Thinking)
36.7%
#52
Grok 3 (Thinking)
36.7%
#53
OpenAI o1 Mini
33.3%
#54
Mistral Large 2
32.7%
#55
GPT-3.5 Turbo
23.3%
#56
Cohere Command A
22.7%
#57
Cohere Command R+
20.0%