Anthropic Model-Written Evaluations
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
Anthropic Model-Written Evaluations (MWE) In-distribution
2.363Steerability
8
Anthropic Model-Written Evaluations Sycophancy (test)
70Average Token Probability
7
Anthropic Model-Written Evaluations Survival Instinct (test)
93Average Token Probability
7
Anthropic Model-Written Evaluations Myopic Reward (test)
0.99Average Token Probability
7
Anthropic Model-Written Evaluations Hallucination (test)
39Avg Token Probability
7
Anthropic Model-Written Evaluations Corrigibility (test)
94Average Token Probability
7
Anthropic Model-Written Evaluations AI Coordination (test)
34Average Token Probability
7