Ai2 Releases Open-Weight Visual Web Agent MolmoWeb

Ai2 has launched MolmoWeb, an open-weight visual web agent designed to assist engineers building browser automation tools. This new offering provides a fully trained model with a comprehensive training stack and a large dataset of human task trajectories, addressing a gap between closed APIs and un-trained open frameworks. MolmoWeb operates using browser screenshots, enabling it to perform actions directly on web pages without relying on HTML parsing. The accompanying dataset, MolmoWebMix, includes 30,000 human task trajectories and millions of question-answer pairs, making it the largest publicly available collection of its kind. This release aims to empower developers with a transparent, auditable, and customizable solution for AI-driven web interaction.