ARM AI

AI Models

Comprehensive risk assessments for 8 frontier AI models

MODEL
COMPANY
VERSIONACCESS
RISK SCORE
OpenAI GPT-5 Nano
Most capable OpenAI model with strongest performance on complex tasks
OpenAI		APIAPI
26.0Moderate
Anthropic Claude 3.5 Opus
Most capable Anthropic model with strongest performance on complex tasks
Anthropic		APIAPI
28.0Moderate
Google Gemini 2.5 Flash Lite
Most capable Google model with strongest performance on complex tasks
Google		APIAPI
33.0Moderate
xAI Grok 4 Fast
Most capable xAI model with strongest performance on complex tasks
xAI		APIAPI
39.0Moderate
Meta Llama 4 70B
Most capable Meta model with strongest performance on complex tasks
Meta		APIAPI
42.0Moderate
DeepSeek V3
Most capable DeepSeek model with strongest performance on complex tasks
DeepSeek		APIAPI
45.0Moderate
Mistral Large 2
Most capable Mistral model with strongest performance on complex tasks
Mistral		APIAPI
47.0High Risk
Moonshot K2
Most capable Moonshot AI model with strongest performance on complex tasks
Moonshot AI		APIAPI
50.0High Risk