MedFlowBench (Radiology) - Localization: leaderboard
Metric: Localization (0-1, shown times 100): share of cases whose reported RAS point lies inside the hidden lesion mask, averaged over the brain MRI and lung PET/CT modules; the model drives 3D Slicer through the MedOpenClaw runtime with viewer-native actions only (series selection, scrolling, windowing, prealigned fusion), inspecting the complete study within a 20-round budget and returning an answer with structured evidence (Track A); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 59 | #54 |
| 2 | Gemini 3 Flash (Preview) | 57 | #78 |
| 3 | GPT-5.5 | 49 | #26 |
| 4 | Claude Opus 4.7 | 35 | #45 |
| 5 | GPT-5.4 Mini | 19 | #202 |
| 6 | Claude Sonnet 4.6 | 18 | #85 |
| 7 | Qwen 3.5 27B | 18 | #183 |
| 8 | Qwen 3.5 35B A3B | 6 | #250 |
| 9 | Qwen 3.5 9B | 4 | #363 |
| 10 | Gemma 3 27B (IT) | 3 | #509 |
| 11 | Qwen 3.5 4B | 2 | #470 |
| 12 | MedGemma 1.5 4B | 0 | #775 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=medflowbench-radiology-localization · How It Works · Data refreshed daily, snapshot 2026-10-11.