Open data and scalable training for long-horizon video world models.
Not enough history yet to draw a chart.