CareQA-Vision - Open-Ended: leaderboard
Image-based questions from the 2020-2024 Spanish MIR and EIR residency entrance exams, rewritten as open-ended questions and checked by hand; CareQA-Vision (Guasch-Martí et al., Barcelona Supercomputing Center, 2026). Each of the 136 answers is scored 0, 0.5 or 1 against the exam answer by the majority of three open LLM judges, and the mean is given in percent; higher is better.
Metric: Judge score (%; 136 open-ended image-based questions from MIR and EIR exams, majority of three LLM judges scoring 0, 0.5 or 1). Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 74.63 | #172 |
| 2 | GLM-4.5V | 64.71 | #739 |
| 3 | Qwen 2 VL 72B Instruct | 45.59 | #1403 |
| 4 | Qwen 3 VL 30B A3B Instruct | 43.75 | #870 |
| 5 | Qwen 3 VL 8B Instruct | 38.24 | #982 |
| 6 | Qwen 2 VL 7B Instruct | 26.47 | #2569 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=careqa-vision-open-ended · How It Works · Data refreshed daily, snapshot 2026-11-12.