CaReCoS: leaderboard
CaReCoS (Centific, 2026) asks multimodal models clinically grounded questions about heart, lung and cough recordings from seven public medical audio datasets, shown only as mel-spectrogram images. An LLM judge scores each answer against the reference; the score is accuracy in percent weighted over all 5,040 questions, higher is better.
Metric: Accuracy (%) of one-to-two-sentence answers to clinically grounded questions about a mel-spectrogram image of the recording, judged by an LLM against the reference (strict factual matching for explicit questions, relaxed semantic matching for inferred ones); question-weighted average over all 5,040 questions from seven datasets; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 | 48.8 | #229 |
| 2 | Gemini 2.5 Pro | 43.2 | #266 |
| 3 | Gemini 2.5 Flash | 24.8 | #453 |
| 4 | Qwen2.5-Omni-7B | 18.4 | #1789 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=carecos · How It Works · Data refreshed daily, snapshot 2026-11-13.