Open Models Score in Top 600 to 3000 for Spanish MIR Medical Exam
October 23, 2026
Four open language models sized between 8B and 27B parameters were evaluated on the Spanish MIR medical residency exam. Results show that while models can rank among the top candidates, they struggle with uncertainty, failing to abstain even when confidence is low, and show varying ability to utilize medical images.
HOW THIS AFFECTS YOU
●
researcherNote that medical fine-tuning is required to mitigate overconfidence in small-scale open models.
●
healthBe aware that current open models lack reliable uncertainty calibration for clinical decision support.