OpenMHC releases decade-long wearable health dataset with open foundation model weights
Researchers behind the My Heart Counts study introduce OpenMHC, a large wearable health dataset for qualified researchers, plus open-source code and weights for wearable foundation models and a standardized benchmark.
In plain words
A team has published OpenMHC, a large collection of sensor and health data from wearable devices, together with openly reusable code and model weights. The goal is to let researchers reproduce and compare wearable health AI, since the biggest existing datasets and models are closed. Their benchmark shows tuned classical methods still rival foundation models.
What was announced
The authors describe OpenMHC as the largest and most comprehensive wearable health dataset broadly accessible to qualified researchers, built from more than a decade of data collected through the My Heart Counts study app. It contains over 67 million participant-hours of wearable recordings from 11,894 participants, with up to 19 sensor channels such as step count, heart rate, sleep and workouts, and up to 170 health, lifestyle, mood and behavioral variables. The release also includes open-source implementations and weights for recent wearable foundation models, which the authors note are rarely open weight or reproducible when trained on proprietary data.
To standardize evaluation, the team established a benchmark covering three tasks: health and behavioral outcome prediction, multivariate data imputation, and time-series forecasting. Benchmarking classical methods alongside recent wearable and time-series foundation models, they report that foundation models perform strongly, but carefully tuned classical baselines such as XGBoost remain highly competitive. The authors position the combination of broad-access data, open code and open weights as a framework for reproducible advances in wearable health AI.
What it does not show
- The evidence is the paper's abstract only; the full article's methods, licensing terms and access conditions were not reviewed, and 'broadly accessible to qualified researchers' is not further defined in the excerpt.
- Performance findings are the authors' own benchmark results and have not been independently confirmed.
For security and IT teams
- Access
- The dataset is described as broadly accessible to qualified researchers, suggesting an application or qualification step rather than unrestricted download.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.