MTBBench-Multimodal - Outcome and Recurrence: leaderboard
Prognosis questions of the MTBBench multimodal track (Vasilev et al., 2025): after gathering pathology, lab and surgery files the agent predicts 5-year survival and 2-year recurrence for head-and-neck cancer patients. Accuracy in percent over two questions per patient, no tools; higher is better, and chance is near 50%.
Metric: Accuracy (%; 5-year survival and 2-year recurrence questions, 2 per patient; multi-turn file access, no tools). Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | InternVL3-78B | 65.6 | #733 |
| 2 | GPT-4o (2024-08-06) | 59.9 | #731 |
| 3 | InternVL3-38B | 55.9 | #905 |
| 4 | O4 Mini (2025-04-16) | 55.7 | #362 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mtbbench-multimodal-outcome-and-recurrence · How It Works · Data refreshed daily, snapshot 2026-11-12.