Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-TypesOfDefamation

A measure of model refusal for Defamation (Level-1: Societal Risks, Level-2: Defamation) related to types of defamation. Includes Level-4 risks like disparagement, libel, and slander.
Source:

Model Performance

#1
Claude-3.5-Sonnet-1022
100.0%
#2
Claude-4.0-Sonnet
99.1%
#3
Claude-4.0-Opus-Thinking
99.1%
#4
Claude-4.0-Opus
99.1%
#5
Claude-4.0-Sonnet-Thinking
99.1%
#6
GPT-OSS-120B
96.3%
#7
Claude-3-Opus
92.6%
#8
Claude-3-Sonnet
91.7%
#9
Claude-3.7-Sonnet-Thinking
91.7%
#10
Claude-3.7-Sonnet
91.7%
#11
Granite-3.0
91.7%
#12
OpenAI-O1-1217
88.0%
#13
OpenAI-O3-medium
88.0%
#14
GPT-4.5
85.2%
#15
Kimi-K2-Instruct
80.6%
#16
OpenAI-O4-mini-medium
80.6%
#17
OpenAI-O3-mini-medium
77.8%
#18
Gemini-2.5-Pro-Thinking
73.1%
#19
Gemini-2.5-Pro-0325
73.1%
#20
GPT-4.1
69.4%
#21
Gemini-2.5-Flash
65.7%
#22
Llama-4-Maverick-17B
63.9%
#23
Grok-3-Mini-Beta
60.2%
#24
GPT-4o-0513
60.2%
#25
Qwen-3
60.2%
#26
Qwen-3-Thinking
60.2%
#27
DeepSeek-R1
59.3%
#28
Gemini-2.0-Flash
58.3%
#29
Gemini-1.5-Pro
57.4%
#30
Grok-4
55.6%
#31
Grok-4-Thinking
55.6%
#32
Gemini-2.0-Pro-0121
54.6%
#33
GPT-3.5-Turbo
53.7%
#34
Grok-3-Beta
51.9%
#35
GPT-4-mini
50.9%
#36
Llama-3.1-405B
50.9%
#37
OpenAI-O1-mini
43.5%
#38
DeepSeek-V3
38.9%
#39
Mistral-Large-2
37.0%
#40
Cohere-Command-R-Plus
22.2%