While access to basic education has increased greatly in Ethiopia and funding for education has improved notably, learning levels among pupils do not appear to have improved commensurably and have likely deteriorated. We explore in this paper the trend in mathematics learning in relation to the General Education Quality Improvement Programme - Phase II (GEQIP-II). We make use of a unique longitudinal dataset on 33 schools in six regions of Ethiopia covering the period 2012 to 2018. We found that pupils’ progress in mathematics in the 2018-19 academic year improved slightly compared to 2012-13, but there is a difference in magnitude of learning progress for the two periods between pupils across rural-urban locations, regional states, and family economic backgrounds. There is an overall improvement in measures of school infrastructure and in teacher qualifications between 2012 and 2018, and there is evidence of changes in student composition between the two periods. Consistent with the GEQIP-II reforms in terms of supporting access and retention, pupils in 2018-19 were more likely to have attended pre-school, have lower absence rates, and have fewer episodes of dropout compared to pupils in the same grade in 2012-13. Compared to pupils in 2012, those in the 2018 cohort had caregivers that were less likely to be literate, and had fewer assets at home. Differences in mathematics learning levels and learning progress between disadvantaged pupils and their relatively advantaged counterparts are discussed in relation to the GEQIP-II reforms.