Abstract The widely held assumption that international collaboration boosts scientific impact oversimplifies a more complex relationship. This study examines how different levels of international collaboration affect the citation impact of the most productive countries and research fields, based on Scopus data from 1996 to 2017. Both field-normalized and non-normalized citation metrics are employed. While international collaboration and scientific impact are highly correlated under standard linear regression, such models fail to capture the skewed and heteroskedastic nature of citation distributions. To address this limitation, we apply pooled quantile regression with panel data covering 23 countries. Three key findings emerge: (i) Higher levels of international collaboration are generally associated with increased citation impact across quantiles. However, this effect varies substantially across disciplines, even among highly productive fields such as Medicine, Physics, Engineering, and Computer Science, suggesting that field-specific citation practices, rather than collaboration per se, may drive the observed differences; (ii) When using field-normalized metrics, the increasing effect of collaboration across quantiles largely disappears; (iii) Significant regional disparities emerge when controlling for geography through dummy variables. These findings challenge the validity of one-size-fits-all collaboration policies and underscore the importance of tailoring science strategies to each country’s research capacity and disciplinary strengths. The study offers actionable insights for evidence-based science and innovation policymaking.