Accuracy and stability of DeepSeek-R1 and GPT-4o on pediatric questions - Takeaways - MDSpire

Accuracy and stability of DeepSeek-R1 and GPT-4o on pediatric questions

  • By

  • Qibo Hu

  • Lin Lei

  • Xuechun Wang

  • Fangshu Liu

  • Guanghua Che

  • July 17, 2026

Share

  • 1

    DeepSeek-R1 and GPT-4o were evaluated using 280 pediatric licensing exam questions across 11 subspecialties.

  • 2

    DeepSeek-R1 outperformed GPT-4o in per-run accuracy, aggregate accuracy, and consistent accuracy across both runs.

  • 3

    DeepSeek-R1 achieved per-run accuracy rates of 90.7% and 86.8%, while GPT-4o had rates of 81.8% and 79.3%.

  • 4

    Both models showed comparable inter-run consistency, with DeepSeek-R1 at 91.4% and GPT-4o at 87.9%.

  • 5

    Performance declined with increasing question complexity, affecting both models' accuracy on more complex case scenarios.

Original Source(s)

Accuracy and stability of DeepSeek-R1 and GPT-4o on pediatric questions

  • by Qibo Hu, Lin Lei, Xuechun Wang, Fangshu Liu, Guanghua Che

  • July 17, 2026

Related Content

The relationship between post-traumatic sleep and related symptoms in children with high-energy trauma: a study based on ecological momentary assessment

Influencing factors of peripheral blood indicators on 30-day mortality risk in children with hemophagocytic lymphohistiocytosis based on weighted quantile sum regression: a single-center retrospective cohort study

Medical Oddities: Natural Doesn't Always Mean Safe

From the gym to the kitchen to romance, these studies challenge everyday assumptions.