WearableQA - Overall (Direct Answer): 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. Overall accuracy when the model must reply with the option letter only, without reasoning, chance 10% (Table 7). Higher is better.
Metric: Accuracy (%; 10-option multiple choice, direct answer). Source: arxiv.org. Saturation forecast: Around February 2028. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 71.3 | #64 |
| 2 | Claude Sonnet 4 | 48.3 | #207 |
| 3 | Claude Opus 4.6 | 40.6 | #67 |
| 4 | Gemma 4 26B A4B | 35.7 | #360 |
| 5 | GPT-5.4 | 33.5 | #83 |
| 6 | Gemini 2.5 Pro | 32 | #159 |
| 7 | Llama 3.3 70B | 27.2 | #578 |
| 8 | Mistral Small 3.1 | 24.5 | #623 |
| 9 | GPT-4o | 23.9 | #357 |
| 10 | Gemma 3 27B | 23.8 | #606 |
| 11 | Llama 3.1 8B | 22.3 | #1130 |
| 12 | Gemma 3 12B | 21.7 | #674 |
| 13 | Gemma 3 4B | 20.6 | #1088 |
| 14 | Llama 3.2 3B | 17.9 | #1376 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=wearableqa-overall-direct-answer · How It Works · Data refreshed daily, snapshot 2026-11-05.