WearableQA - Health Reasoning: leaderboard
WearableQA (arXiv 2609.05405v1): 4,084 ten-option multiple-choice questions over wearable time series, blood biomarkers and demographics of 200 real users with up to 500 days of data. Accuracy on health-reasoning questions that need clinical and physiological interpretation, chain-of-thought (Table 1). Higher is better.
Metric: Accuracy (%; 10-option multiple choice, chain-of-thought). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 67.8 | #64 |
| 2 | Claude Opus 4.6 | 66.3 | #67 |
| 3 | Claude Sonnet 4 | 65.2 | #207 |
| 4 | GPT-5.4 | 63.5 | #83 |
| 5 | Gemma 4 26B A4B | 59.8 | #360 |
| 6 | Gemini 2.5 Pro | 56.8 | #159 |
| 7 | GPT-4o | 53.5 | #357 |
| 8 | Mistral Small 3.1 | 47.9 | #623 |
| 9 | Gemma 3 27B | 40.7 | #606 |
| 10 | Llama 3.3 70B | 40.7 | #578 |
| 11 | Gemma 3 12B | 39.9 | #674 |
| 12 | Llama 3.1 8B | 36.8 | #1130 |
| 13 | Llama 3.2 3B | 31.7 | #1376 |
| 14 | Gemma 3 4B | 29.4 | #1088 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=wearableqa-health-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-05.