MedMosaic - Speech MCQ: leaderboard
MedMosaic (Rajgarhia et al., 2026): 11,754 adversarial multiple-choice questions on short clinical speech whose answers depend on what is audible. Accuracy in percent (higher is better).
Metric: Accuracy (%; multiple choice, extracted answer letter against the answer key; 11,754 questions on short clinical speech from Primock, Ekacare, Hani89, MultiMed and MedDialog-Audio). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 68 | #262 |
| 2 | Gemini 2.5 Flash | 66.6 | #452 |
| 3 | Qwen2.5-Omni-7B | 52.7 | #1797 |
| 4 | Gemma 3n E4B (IT) | 42.8 | #2220 |
| 5 | Phi-4 Multimodal Instruct | 32.8 | #3914 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medmosaic-speech-mcq · How It Works · Data refreshed daily, snapshot 2026-11-12.