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
Claude-3.5-Sonnet-1022
86.2%
#2
Claude-4.0-Sonnet
84.8%
#3
Claude-4.0-Sonnet-Thinking
84.8%
#4
GPT-OSS-120B
82.6%
#5
Claude-4.0-Opus
79.0%
#6
Claude-4.0-Opus-Thinking
79.0%
#7
Granite-3.0
69.6%
#8
Claude-3.7-Sonnet
65.9%
#9
Claude-3.7-Sonnet-Thinking
65.9%
#10
Claude-3-Opus
64.5%
#11
Claude-3-Sonnet
61.6%
#12
Gemini-2.0-Pro-0121
61.6%
#13
Llama-4-Maverick-17B
61.6%
#14
OpenAI-O3-medium
61.6%
#15
Kimi-K2-Instruct
60.9%
#16
Gemini-2.5-Pro-0325
59.4%
#17
Gemini-2.0-Flash
59.4%
#18
Gemini-2.5-Pro-Thinking
59.4%
#19
Gemini-1.5-Pro
55.8%
#20
OpenAI-O4-mini-medium
55.1%
#21
Gemini-2.5-Flash
55.1%
#22
DeepSeek-R1
55.1%
#23
OpenAI-O1-1217
52.9%
#24
OpenAI-O3-mini-medium
52.2%
#25
GPT-4.1
50.7%
#26
GPT-4.5
50.0%
#27
Qwen-3-Thinking
50.0%
#28
Qwen-3
50.0%
#29
Llama-3.1-405B
47.8%
#30
Grok-3-Mini-Beta
47.8%
#31
GPT-4o-0513
44.2%
#32
GPT-4-mini
42.0%
#33
Grok-3-Beta
40.6%
#34
OpenAI-O1-mini
39.1%
#35
DeepSeek-V3
39.1%
#36
Grok-4
37.7%
#37
Grok-4-Thinking
37.7%
#38
GPT-3.5-Turbo
37.0%
#39
Mistral-Large-2
36.2%
#40
Cohere-Command-R-Plus
30.4%