Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-InfluencingPolitics

A measure of model refusal for Political Usage (Level-1: Societal Risks, Level-2: Political Usage) related to influencing politics. Includes Level-4 risks like influencing political decisions and influencing political opinions.
Source:

Model Performance

#1
Claude 3 Sonnet
100.0%
#2
Claude 3.5 Sonnet
100.0%
#3
Claude 3.5 Haiku
100.0%
#4
Claude Haiku 4.5 (Thinking)
100.0%
#5
GPT-4.5
100.0%
#6
GPT-5
100.0%
#7
GPT-5 (Thinking)
100.0%
#8
Claude Haiku 4.5
100.0%
#9
GPT OSS 120B
100.0%
#10
Gemini 2.5 Pro
100.0%
#11
Gemini 2.5 Pro (Thinking)
100.0%
#12
Grok 3 Mini
100.0%
#13
Kimi K2
96.7%
#14
Claude 3.7 Sonnet
96.7%
#15
Claude 4.0 Opus (Thinking)
96.7%
#16
Claude 4.0 Opus
96.7%
#17
Claude 3.7 Sonnet (Thinking)
96.7%
#18
Gemini 2.5 Flash
95.6%
#19
Gemini 2.5 Flash (Thinking)
95.6%
#20
OpenAI o4 Mini (Medium Effort)
93.3%
#21
OpenAI o3 Mini (Medium Effort)
93.3%
#22
OpenAI o3 Mini (High Effort)
93.3%
#23
OpenAI o4 Mini (High Effort)
93.3%
#24
OpenAI o1
92.9%
#25
Claude 3 Opus
90.0%
#26
Gemini 2.0 Flash (Thinking)
90.0%
#27
Granite 3.0
90.0%
#28
GPT-4.1
90.0%
#29
GPT-4o Mini
90.0%
#30
Gemini 2.0 Flash Thinking (Jan 2025)
90.0%
#31
Gemini 2.0 Flash
90.0%
#32
OpenAI o3 (Medium Effort)
86.7%
#33
OpenAI o3 (High Effort)
86.7%
#34
Qwen 3
80.0%
#35
Qwen 3 (Thinking)
80.0%
#36
Claude Sonnet 4.5
77.0%
#37
Claude Sonnet 4.5 (Thinking)
77.0%
#38
Llama 3.1 405B
76.7%
#39
Grok 3 (Thinking)
73.3%
#40
Grok 3
73.3%
#41
GPT-4o
73.3%
#42
DeepSeek R1
71.7%
#43
Gemini 2.0 Pro
68.3%
#44
GPT-3.5 Turbo
63.3%
#45
Llama 4 Maverick 17B
57.6%
#46
Claude 4.0 Sonnet
56.7%
#47
Claude 4.0 Sonnet (Thinking)
56.7%
#48
OpenAI o1 Mini
50.0%
#49
Gemini 1.5 Pro
43.8%
#50
DeepSeek V3
43.3%
#51
Llama 3.3 70B
43.3%
#52
Gemini 3.1 Pro Preview
34.0%
#53
Cohere Command A
23.3%
#54
Cohere Command R+
20.0%
#55
Grok 4
16.7%
#56
Mistral Large 2
16.7%
#57
Grok 4 (Thinking)
16.7%