Previous randomized studies have shown that addressing children’s current learning gaps, rather than following an over-ambitious uniform curriculum, can lead to significant learning gains. In this study, we evaluate a series of efforts to scale up the NGO Pratham’s approach to teaching children according to their actual learning level, in four Indian States. While this approach was previously shown to be extremely effective when implemented with community volunteers outside of school, the objective of these new scale-up evaluations was to develop a model that could be implemented within the government school system. In the first two instances (Bihar and Uttarakhand), the methodology was not adopted by government schoolteachers, despite well-received training sessions and Pratham support. Motivated by the quantitative and qualitative analysis of these early attempts, we adapted the approach and designed large-scale experiments in the states of Haryana and Uttar Pradesh to test two new scale-up models. In Haryana, teachers received support from government resource persons trained by Pratham, and implemented the approach during a dedicated hour. In Uttar Pradesh, Pratham volunteers implemented high-intensity, short-burst “learning camps” for 40 days, in school and during school hours, with additional 10-day summer camps. Both models proved effective, with gains in language of 0.15 standard deviation in Haryana, and 0.70 standard deviations in Uttar Pradesh, on all students enrolled in these schools at baseline. These two models provide blueprints that can be replicated inside other government systems.