FUTO releases a 1M-swipe English typing dataset used to train and evaluate its new swipe keyboard model

FUTO collected more than 1 million voluntary QWERTY English swipe samples through swipe.futo.org, asking mobile users to swipe sentence prompts word by word. After filtering low-quality entries, the organization released the dataset on HuggingFace under the MIT license and used it to train and benchmark swipe typing models.

Open swipe-typing datasets are relatively rare, so this release could help independent developers build better mobile keyboards without relying on proprietary data. The MIT licensing also points to a broader push for open-source input methods that can compete with closed keyboard ecosystems on accuracy and privacy.