MedFrameQA: leaderboard
MedFrameQA is a multi-image medical VQA benchmark for clinical reasoning over diagnostic sequences. It contains 2,851 high-quality multi-image VQA pairs with transcript-grounded reasoning chains, drawn from 9,237 frames in 3,420 medical education videos, and tests whether MLLMs can synthesize two to five images rather than treating each frame in isolation. (Source: benchmarklist.com, self-reported
Metric: Average Accuracy (self-reported). Source: benchmarklist.com. Saturation forecast: Around October 2026. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 54.75 | #245 |
| 2 | O3 | 50.18 | #124 |
| 3 | Claude 3.7 Sonnet | 49.67 | #249 |
| 4 | O4 Mini | 49.4 | #180 |
| 5 | O1 | 47.91 | #183 |
| 6 | GPT-4 Turbo | 46.69 | #404 |
| 7 | QVQ-72B-Preview | 46.44 | #440 |
| 8 | GPT-4o | 45.67 | #343 |
| 9 | Qwen 2.5 VL 72B Instruct | 42.65 | #377 |
| 10 | GPT-4o Mini | 34.55 | #599 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=medframeqa · How It Works · Data refreshed daily, snapshot 2026-10-20.