Studies on collaborative dialogue, operationalized as language-related episodes (LREs), across task modalities (oral vs. oral+written) have received limited attention in the literature and the vast majority have examined adult learners (Adams & Ross-Feldman, 2008; Niu, 2009; Payant & Kim, 2017), except for García Mayo and Imaz Agirre (2019) with young learners. In addition, all these studies have ascribed the greater incidence and higher number of resolved LREs in oral+written tasks to the extra processing time learners have to reflect on their production. However, no studies exist that have actually tested whether asking learners to attend to accuracy in both modalities would yield similar results. Therefore, this study analyses the number and types of LREs produced by primary-school Basque/Spanish bilinguals learning L3 English in a CLIL setting. 23 proficiency-matched dyads were asked to work together and attend to accuracy while completing two tasks with different modalities: an oral task and an oral+written task. Results indicate that in terms of quantity, the oral+written mode promoted a greater number of LREs and in particular, more form-focused LREs. As for types, meaning-focused LREs were far more frequent than form-focused LREs in the oral task, while no differences emerged in the oral+written task. In addition, the accurate resolution of LREs surpassed the production of unresolved and non target-like resolved LREs only in the oral+written mode. These findings support the superiority of the written mode in the promotion of learning opportunities for young learners, despite having been encouraged to focus on form in the oral modality.