{"data":{"id":1603,"lang":"en","title":"OpenAI o1 Model Beats ER Physicians at Diagnosis With 67% Accuracy in Boston Trial, Science Journal Finds","summary":"OpenAI's o1 reasoning model beat hundreds of physicians at diagnosis in a Science journal study, hitting 67.1% triage accuracy vs 55.3% and 50% for attending doctors in Boston ER cases.","category":"science","category_label":"Science","published_at":"2026-05-07T06:01:41Z","updated_at":"2026-05-07T06:01:41Z","url":"https://techtools.cz/tools/news/article/1603-openai-o1-model-beats-er-physicians-at-diagnosis-with-67-accuracy-in-boston","api_url":"https://techtools.cz/news-api/v1/articles/1603?lang=en","markdown_url":"https://techtools.cz/news-api/v1/articles/1603.md?lang=en","image_url":"https://www.statnews.com/wp-content/uploads/2026/04/s3___bgmp-arc_arc-feeds_generic-photos_to-arc_ryanhospital1met-645x645.jpg","word_count":156,"reading_time_minutes":1,"sources":[{"url":"https://www.science.org/doi/10.1126/science.adz4433","domain":"science.org"},{"url":"https://www.statnews.com/2026/04/30/open-ai-llm-model-outperforms-doctors-study-published-journal-science/","domain":"statnews.com"},{"url":"https://www.npr.org/2026/04/30/nx-s1-5804474/ai-doctors-openai-patient-care-diagnosis","domain":"npr.org"}],"translation":{"lang":"cs","id":1676,"title":"Model OpenAI o1 překonal pohotovostní lékaře v diagnostice se 67% přesností v bostonské studii, ukázal časopis Science","url":"https://techtools.cz/tools/news/clanek/1676-model-openai-o1-prekonal-pohotovostni-lekare-v-diagnostice-se-67-presnosti-v","api_url":"https://techtools.cz/news-api/v1/articles/1676?lang=cs"},"content":"A study published in the journal Science found that OpenAI's o1 reasoning model outperformed hundreds of physicians at diagnostic and clinical reasoning tasks. Led by Beth Israel Deaconess internist Adam Rodman, the trial pitted the model against doctors on both published clinicopathological cases and live emergency department patients in Boston.\n\nOn real ER triage cases, o1 reached exact or very close diagnostic accuracy in 67.1% of patients, beating two attending physicians who scored 55.3% and 50.0%. On published case vignettes, o1-preview hit 88.6% accuracy, well ahead of GPT-4's 72.9% and the prior generation of clinical decision support tools.\n\nRodman cautioned the results should not be read as proof the model is safe to deploy on patients, noting the AI worked from text alone and never sees the images, sounds, or nonverbal cues that clinicians rely on. The paper explicitly answers a 1959 Science article that defined the threshold for an AI system surpassing human diagnostic reasoning."}}