Spanish learners engaged in two-way interaction gap tasks where, through social interaction, they identified and defined low-frequency vocabulary. Participants (referred to as actors to differentiate them from task partners) completed two such tasks. Each time, they collaborated with different partners of varying degrees of Spanish proficiency and linguistic background, either an L2 learner or a heritage speaker (HS). The purpose of the study was thus to shed light on if and how features of the collaborating partners in mixed (HS-L2) and matched (L2-L2; HS-HS) partnerships impacted vocabulary learning. Through a mixed methods analysis of the data (perception measures and qualitative responses, plus pre- and post-vocabulary tests), it was deduced that learners' language proficiency level alone was not associated with differences in vocabulary gains. Rather, the nature of the partnerships among learners was more salient. In particular, learners in mixed partnerships had greater gains than those in matched partnerships. In addition, actors' perceptions of the experience of working with their partner had no discernible effect on vocabulary, except for HSs who increased less when working with partners' whose linguistic abilities they had rated as low. Pedagogical implications address learners' social interaction in mixed language classrooms.