Longitudinal MRI Progression Reasoning: leaderboard
Final disease-progression accuracy for 16 general and medically specialized vision-language models on zero-shot multi-sequence, multi-view longitudinal MRI inputs. This is the main benchmark evaluation, separate from the UCSF-GBM agentic-workflow subset.
Metric: Final Accuracy (%, zero-shot multi-view MRI). Source: arxiv.org. Saturation forecast: Not forecast. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 35.1 | #80 |
| 2 | O4 Mini | 32.18 | #180 |
| 3 | GPT-4o | 32 | #343 |
| 4 | Llama 4 Scout | 28.47 | #656 |
| 5 | Llama 4 Maverick | 26.84 | #464 |
| 6 | Qwen 3 VL 8B Instruct | 24.31 | #420 |
| 7 | Gemini 2.5 Pro | 23.66 | #152 |
| 8 | Gemini 2.5 Flash | 23.57 | #245 |
| 9 | MedGemma-4B-IT | 23.5 | #853 |
| 10 | Gemini 3 Flash | 22.3 | #99 |
| 11 | GPT-5.2 | 21.2 | #110 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=longitudinal-mri-progression-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.