MedRoundsQA - Multi-Turn Diagnosis: leaderboard
MedRoundsQA diagnosis accuracy in percent, preserving the source doctor-agent checkpoint identifiers in Appendix Table 9. Single-turn uses standardized record-style input; multi-turn scores the interactive diagnostic consultation. Reported means retain separate model identities.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around 2029. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 41.92 | #152 |
| 2 | MiniMax-M2.5 | 36.4 | #316 |
| 3 | GPT-4o (2024-08-06) | 31.55 | #343 |
| 4 | GLM-5 | 31.21 | #149 |
| 5 | Qwen 3 VL 32B Instruct | 28.77 | #290 |
| 6 | DeepSeek V3 | 27.14 | #329 |
| 7 | Gemini 2.5 Flash (Preview 05-20) | 26.68 | #229 |
| 8 | Llama 3 8B UltraMedical | 22.21 | #1186 |
| 9 | Ministral-3-8B-Instruct-2512 | 19.92 | #663 |
| 10 | Phi-4 | 19.78 | #707 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=medroundsqa-multi-turn-diagnosis · How It Works · Data refreshed daily, snapshot 2026-11-05.