Llama 3.3 70B Instruct
Meta
Alignment score: 85 out of 100High alignment25 facts
Language Models
14 leading models from different providers, analyzed for alignment and evasion patterns.
14 models · 25 facts
United States
6 modelsGPT-4o
OpenAI
Alignment score: 84 out of 100High alignment25 facts
Gemini 2.5 Pro
Alignment score: 83 out of 100High alignment25 facts
Claude Opus 4.7
Anthropic
Alignment score: 82 out of 100High alignment25 facts
Grok 4.3
xAI
Alignment score: 81 out of 100High alignment25 facts
Nous Hermes 4 70B
Nous Research
Alignment score: 72 out of 100Mixed alignment25 facts
Europe
preliminary · n=11 modelChina
5 modelsQwen 3
Alibaba
Alignment score: 87 out of 100High alignment25 facts
Seed 2.0 Lite
ByteDance
Alignment score: 86 out of 100High alignment25 facts
GLM-4.7
Zhipu AI
Alignment score: 86 out of 100High alignment25 facts
DeepSeek R1
DeepSeek
Alignment score: 83 out of 100High alignment25 facts
Kimi K2.6
Moonshot AI
Alignment score: 80 out of 100High alignment25 facts