Evaluation of 47 leading AI models using the Tandem harness for medical note generation based on 50 transcripts selected to be representative across European languages and care settings. The models were scored on clinician preference and accuracy. Clinician preference measures how often the model’s output medical note was preferred versus the output medical note of other models evaluated in the benchmark. Accuracy measures the share of clinical claims in the output medical notes that were neither hallucinations nor omissions, with omissions being measured on a defined set of clinically important facts that were contained in the transcripts that should be included in the resulting note.