Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

AIR-Bench-HighRiskFinancialActivities

A measure of model refusal for Economic Harm (Level-1: Societal Risks, Level-2: Economic Harm) related to high-risk financial activities. Includes Level-4 risks like gambling (e.g., sports betting) and payday lending.

Model Performance

#1
Claude-3-Sonnet
100.0%
#2
Claude-3.5-Sonnet-1022
100.0%
#3
Granite-3.0
100.0%
#4
GPT-OSS-120B
100.0%
#5
OpenAI-O3-medium
100.0%
#6
Claude-4.0-Sonnet
96.7%
#7
Claude-3-Opus
96.7%
#8
Claude-4.0-Opus
96.7%
#9
OpenAI-O4-mini-medium
96.7%
#10
OpenAI-O1-1217
96.7%
#11
Claude-4.0-Opus-Thinking
96.7%
#12
Claude-4.0-Sonnet-Thinking
96.7%
#13
Gemini-2.0-Pro-0121
93.3%
#14
Llama-4-Maverick-17B
93.3%
#15
Gemini-2.0-Flash
93.3%
#16
GPT-4.5
93.3%
#17
Claude-3.7-Sonnet-Thinking
90.0%
#18
Claude-3.7-Sonnet
90.0%
#19
OpenAI-O3-mini-medium
86.7%
#20
GPT-4-mini
86.7%
#21
GPT-4.1
80.0%
#22
Kimi-K2-Instruct
80.0%
#23
Gemini-2.5-Pro-0325
76.7%
#24
Gemini-1.5-Pro
76.7%
#25
Gemini-2.5-Flash
76.7%
#26
Gemini-2.5-Pro-Thinking
76.7%
#27
Qwen-3-Thinking
73.3%
#28
Qwen-3
73.3%
#29
Llama-3.1-405B
70.0%
#30
DeepSeek-R1
66.7%
#31
GPT-4o-0513
66.7%
#32
GPT-3.5-Turbo
66.7%
#33
Grok-3-Beta
65.0%
#34
Grok-3-Mini-Beta
46.7%
#35
DeepSeek-V3
43.3%
#36
Grok-4-Thinking
36.7%
#37
Grok-4
36.7%
#38
OpenAI-O1-mini
33.3%
#39
Mistral-Large-2
33.3%
#40
Cohere-Command-R-Plus
20.0%