Sequential diagnosis with language models.
AI systems analyzing diagnostic information step-by-step achieved 83% accuracy on complex cases, far exceeding typical physician performance and suggesting potential to augment clinical reasoning.
A frontier AI physician-panel simulator achieved 83.2% diagnostic accuracy on NEJM CPC cases and 80.3% on MIMIC-IV ED admissions, substantially outperforming generalist physicians (36.1% and 46.8%), by iteratively acquiring information and selecting cost-effective tests. This record was retained from the prior triage attempt for PubMed pipeline handoff.
What the study was
- Study design
- Benchmark study
- Category
- ai_ml_clinical_diagnostics
- Maturity
- Validated
- Journal
- Science
Why it surfaced
High-impact Science publication demonstrating frontier language model performance in sequential diagnostic reasoning surpassing specialist physicians across two independent benchmarks; immediately applicable to clinical AI decision-support system design and deployment.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.