Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-CelebratingSuffering

A measure of model refusal for Violence & Extremism (Level-1: Content Safety Risks, Level-2: Violence & Extremism) related to celebrating suffering. Includes Level-4 risks like glorifying violence, abuse, or suffering of others; belittling victimhood or violent events; denying well-documented major violent events; and beautifying/whitewashing acts of war or aggression.
Source:

Model Performance

#1
OpenAI o3 (High Effort)
97.8%
#2
OpenAI o3 (Medium Effort)
97.8%
#3
GPT-5 (Thinking)
96.7%
#4
GPT-5
96.7%
#5
Claude 4.0 Sonnet
95.6%
#6
Claude 4.0 Sonnet (Thinking)
95.6%
#7
Claude 4.0 Opus (Thinking)
95.6%
#8
Claude 4.0 Opus
95.6%
#9
Claude 3.5 Haiku
94.4%
#10
OpenAI o1
94.4%
#11
Claude Sonnet 4.5
94.2%
#12
Claude Sonnet 4.5 (Thinking)
94.2%
#13
Gemini 2.5 Pro (Thinking)
88.9%
#14
Gemini 2.5 Pro
88.9%
#15
GPT-4.5
87.8%
#16
Claude 3.5 Sonnet
86.2%
#17
Claude 3.7 Sonnet
65.9%
#18
Claude 3.7 Sonnet (Thinking)
65.9%
#19
Claude 3 Opus
64.5%
#20
Gemini 3.1 Pro Preview
62.2%
#21
Kimi K2
60.9%
#22
Llama 4 Maverick 17B
60.0%
#23
Claude 3 Sonnet
60.0%
#24
Gemini 2.0 Flash
59.4%
#25
Gemini 2.0 Flash Thinking (Jan 2025)
59.4%
#26
Gemini 2.0 Flash (Thinking)
59.4%
#27
Gemini 1.5 Pro
55.8%
#28
DeepSeek R1
55.1%
#29
Gemini 2.5 Flash
55.0%
#30
Gemini 2.5 Flash (Thinking)
55.0%
#31
OpenAI o3 Mini (High Effort)
52.2%
#32
OpenAI o3 Mini (Medium Effort)
52.2%
#33
GPT-4.1
50.7%
#34
Qwen 3
50.0%
#35
Qwen 3 (Thinking)
50.0%
#36
Llama 3.1 405B
47.8%
#37
GPT-4o
44.2%
#38
GPT-4o Mini
42.0%
#39
Claude Haiku 4.5
40.7%
#40
Claude Haiku 4.5 (Thinking)
40.7%
#41
DeepSeek V3
39.1%
#42
OpenAI o1 Mini
39.0%
#43
Llama 3.3 70B
37.7%
#44
Mistral Large 2
36.2%
#45
Cohere Command A
30.4%
#46
Cohere Command R+
30.4%
#47
GPT-3.5 Turbo
26.0%
#48
GPT OSS 120B
22.7%
#49
Grok 4
16.0%
#50
Grok 4 (Thinking)
16.0%
#51
Gemini 2.0 Pro
14.7%
#52
Grok 3 Mini
14.7%
#53
Granite 3.0
8.7%
#54
Grok 3 (Thinking)
6.7%
#55
Grok 3
6.7%
#56
OpenAI o4 Mini (Medium Effort)
4.0%
#57
OpenAI o4 Mini (High Effort)
4.0%