MedMosaic - Voice QA: leaderboard
MedMosaic (Rajgarhia et al., 2026): 1,812 multiple-choice questions whose question and options are spoken inside the audio after the consultation. Accuracy in percent (higher is better).
Metric: Accuracy (%; multiple choice, extracted answer letter against the answer key; 1,812 spoken questions generated from MTS-Dialog). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 70.3 | #262 |
| 2 | Gemini 2.5 Flash | 52.7 | #452 |
| 3 | Phi-4 Multimodal Instruct | 30.6 | #3914 |
| 4 | Qwen2.5-Omni-7B | 29.9 | #1797 |
| 5 | Gemma 3n E4B (IT) | 10.2 | #2220 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medmosaic-voice-qa · How It Works · Data refreshed daily, snapshot 2026-11-12.