This study aims to explore the impact of highly decodable texts (containing only graphemes already taught) versus less decodable ones on the reading performance of first graders. While some specialists recommend the use of solely decodable texts in initial reading instruction (Castles et al., 2018; HMSO, 2005), a recent meta-analysis suggests a blend of decodable and non-decodable texts may be more effective (Pugh et al., 2023). Field research is essential to refine best practices in reading instruction. Reading and comprehension of 246 French children from 15 classrooms were assessed at the beginning of first grade (September) and re-assessed two months later. During this 2-months period, each week, the mean percentage of decodable material read was calculated for each classroom based on the graphemes and the words taught. A multi-level analysis was performed to see whether and how this percentage influences the early acquisition of literacy skills, independently of classroom effect and early abilities effects. Descriptive analyses showed three patterns of decodability: classes using moderately, highly, and increasingly decodable texts. Findings indicated that less decodable texts enhance comprehension, emphasizing the pivotal role of diversity in advancing early literacy skills. The findings challenge the assumption that highly decodable text fosters superior literacy acquisition. Contrarily, it suggests the relevance of incorporating both decodable and non-decodable texts at the onset of reading. Derived from authentic teaching practices, these results hold substantial implications for educational policies and guide educators in refining their instructional approaches.