Claude 3.7 Sonnet
Anthropic
rich detail 20 method fields published
Evaluation
Chain-of-thought faithfulness evaluation
CoT faithfulness score: fraction of prompt pairs (MMLU and GPQA questions with inserted clues) where the model verbalizes the clue as the cause of its answer when the answer changes with the clue
How it was run
Open a release to see the setup, scoring, and other details that the lab published. Only fields the lab actually disclosed are shown.