MolmoWeb - Open web agents from data to deployment
byβ’
Introducing MolmoWeb, an open visual web agent that navigates and completes tasks in a browser using screenshots alone, along with MolmoWebMix, the largest public dataset for training web agents.
Replies
Flowtica Scribe
Hi everyone!
MolmoWeb is one of those rare open releases that really stands out.
It is one of the strongest open web agents out there, able to actually see the screen and act on it. But what is even more impressive is that Ai2 did not stop at open-sourcing the weights β they released the whole stack: training code, eval harness, annotation tooling, synthetic data generation pipeline, client-side demo code, and the full MolmoWebMix dataset.
I honestly do not have much to add beyond respect. Open-sourcing is not an obligation, but doing it at this depth and quality, consistently, is extremely hard. The Ai2 team continues to set a high bar.
This is the kind of foundation the open agent ecosystem actually needs.
screenshots-only is a bold constraint. how does it handle dynamic content that changes between screenshot and click?