Explore official MIR exam questions organized by year. Each question includes detailed analysis of AI model performance.
24 January 2026
Best AI: 198.67 net (Google: Gemini 3 Flash Preview)
Best human: Results pending
25 January 2025
Best AI: 190.67 net (Google: Gemini 3 Flash Preview)
Best human: 165.67 net
20 January 2024
Best AI: 193.67 net (Perplexity: Sonar Deep Research)
Best human: 186.67 net
Evaluation based on zero-shot accuracy on official exam questions.
Puntuación neta: 198.66 pts
OpenAI
Puntuación neta: 198.66 pts
OpenAI
Puntuación neta: 198.66 pts
OpenAI
Puntuación neta: 197.33 pts
OpenAI
Puntuación neta: 197.33 pts
OpenAI
Puntuación neta: 197.33 pts
How we evaluate artificial intelligence models in the medical field using the MIR exam as a reference.
We use real questions from Spain's MIR exam, the standard for evaluating medical knowledge at a professional level. Each question is verified and categorized by specialty.
Each model is evaluated under the same controlled conditions, without access to external information. We measure accuracy, clinical reasoning, and response consistency.
We provide granular metrics by medical specialty, question type, and difficulty level. This allows identifying strengths and areas for improvement for each model.
Distribution of MIR exam questions by subject and type in each edition.
Our benchmark provides an exhaustive evaluation of AI model performance in the medical field.
Performance tracking over time to identify improvements and regressions.
Granular analysis by subject and clinical question type.
Standardized benchmarks based on Spain's official MIR exam.
Open and reproducible methodology with complete access to evaluation criteria.
Periodic incorporation of new models and MIR exam editions.
Rankings and statistics that allow easy performance comparison between models.
Official questions from the Ministry of Health with validated answers.
Question distribution by type
Question distribution by subject
Articles, news and analysis about AI in medicine
The top 33 positions in the MIR 2026 ranking are all proprietary models. We analyze the gap between open and closed models, the real taxonomy of open source in AI, and why RAG outperforms fine-tuning for customizing medical AI.
Read more
Claude Opus 4.6 and GPT-5.2-Codex are the most advanced AI coding models. But in MIR 2026, a Flash model costing just 0.34 euros humiliates them. Analysis of the agentic paradox with data from 290 models.
Read more
Final results of the largest medical AI benchmark in Spanish. Three models tie with 199 correct answers out of 200 valid questions. A 'Flash' model leads for the third consecutive year. Exhaustive analysis of 290 models evaluated with data on cost, speed, tokens, and accuracy.
Read more
