Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-AcademicDishonesty

A measure of model refusal for Deception (Level-1: Societal Risks, Level-2: Deception) related to academic dishonesty. Includes Level-4 risks like plagiarism and promoting academic dishonesty.
Source:

Model Performance

#1
Claude 3.5 Haiku
100.0%
#2
Claude 3.5 Sonnet
100.0%
#3
Claude Sonnet 4.5
100.0%
#4
GPT-5
100.0%
#5
GPT-5 (Thinking)
100.0%
#6
Claude Sonnet 4.5 (Thinking)
100.0%
#7
Claude Haiku 4.5
100.0%
#8
Claude Haiku 4.5 (Thinking)
100.0%
#9
Gemini 3.1 Pro Preview
100.0%
#10
Qwen 3 (Thinking)
100.0%
#11
Qwen 3
100.0%
#12
OpenAI o3 (Medium Effort)
100.0%
#13
OpenAI o3 (High Effort)
100.0%
#14
Claude 4.0 Opus
98.3%
#15
Claude 4.0 Opus (Thinking)
98.3%
#16
Claude 4.0 Sonnet (Thinking)
98.0%
#17
Claude 4.0 Sonnet
98.0%
#18
Claude 3.7 Sonnet (Thinking)
96.7%
#19
Claude 3.7 Sonnet
96.7%
#20
Granite 3.0
96.7%
#21
Claude 3 Opus
96.7%
#22
Claude 3 Sonnet
96.7%
#23
GPT OSS 120B
93.3%
#24
Gemini 2.0 Pro
93.3%
#25
Kimi K2
93.3%
#26
OpenAI o1
90.0%
#27
GPT-4.5
88.9%
#28
Gemini 2.5 Pro
87.5%
#29
Gemini 2.5 Pro (Thinking)
87.5%
#30
GPT-4.1
83.3%
#31
OpenAI o4 Mini (Medium Effort)
77.5%
#32
OpenAI o4 Mini (High Effort)
77.5%
#33
Gemini 2.0 Flash (Thinking)
76.7%
#34
Llama 4 Maverick 17B
76.7%
#35
Gemini 2.0 Flash Thinking (Jan 2025)
76.7%
#36
Gemini 2.0 Flash
76.7%
#37
OpenAI o3 Mini (High Effort)
75.8%
#38
OpenAI o3 Mini (Medium Effort)
75.8%
#39
GPT-4o Mini
73.3%
#40
Gemini 1.5 Pro
70.0%
#41
DeepSeek R1
60.0%
#42
GPT-4o
60.0%
#43
Gemini 2.5 Flash (Thinking)
60.0%
#44
Gemini 2.5 Flash
60.0%
#45
Llama 3.1 405B
58.3%
#46
Grok 3 (Thinking)
40.0%
#47
Grok 3
40.0%
#48
Mistral Large 2
38.3%
#49
Grok 4
36.7%
#50
Llama 3.3 70B
36.7%
#51
Grok 4 (Thinking)
36.7%
#52
OpenAI o1 Mini
33.3%
#53
DeepSeek V3
30.0%
#54
Cohere Command A
23.3%
#55
Grok 3 Mini
23.3%
#56
GPT-3.5 Turbo
18.3%
#57
Cohere Command R+
15.0%