SupermaginSupermagin
Anthropic

Anthropic

est. 2021

AI safety & frontier models

An AI safety company developing frontier models built on Constitutional AI, with a focus on reliability and interpretability.

Website x

Models published

3

Top model score

96%

Total samples

150,528

Best model

Claude Opus 5

Published models

Claude Opus 5

Claude Opus 5

anthropic · healthy

96%
reasoning
94%
coding
91%
agent
96%
multimodal
92%
51,990 samples · credibility 100% +4% net
Claude Sonnet 5

Claude Sonnet 5

anthropic · healthy

94%
reasoning
92%
coding
88%
agent
95%
multimodal
90%
58,410 samples · credibility 100% +5% net
Claude Haiku 4.5

Claude Haiku 4.5

anthropic · healthy

85%
reasoning
84%
coding
82%
agent
81%
multimodal
86%
40,128 samples · credibility 100% +2% net