MTBBench-Longitudinal: leaderboard
The longitudinal track of MTBBench (Vasilev et al., 2025): for 40 MSK-CHORD cancer patients an LLM agent reads clinical timelines, pathology reports and genomic tables that are revealed stage by stage and answers 183 questions on outcome, progression and recurrence, without tools. The score is the unweighted mean of the three task accuracies in percent; higher is better.
Metric: Accuracy (%; unweighted mean of the three task accuracies; 40 patients from MSK-CHORD with clinical timelines and genomic files, 183 true/false and multiple-choice questions; multi-turn agent that requests patient files, no tools). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.3 70B Instruct [4bit] | 66 | #1221 |
| 2 | GPT-4o (2024-08-06) | 64.2 | #731 |
| 3 | O4 Mini (2025-04-16) | 60 | #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-longitudinal · How It Works · Data refreshed daily, snapshot 2026-11-12.