Generational shifts in intelligence test performance, commonly referred to as the Flynn effect, have been the subject of intense empirical investigation. Whilst earlier studies generally had reported systematic gains for full-scale, fluid, and crystallized IQ, more recent findings increasingly indicate heterogeneous change patterns across distinct cognitive domains. Here, we present the first formal meta-analysis of the Flynn effect following the framework of the Cattell–Horn–Carroll (CHC) theory, currently the most widely accepted model of human intelligence. Our quantitative research synthesis integrates more than 3,000 test score changes (1909–2025), based on 2,000,000+ individuals, and including more than 30 CHC stratum I and II domains. Cross-temporal trends were differentiated in terms of sign and magnitude according to specific stratum I domains, with annual changes ranging from –0.11 to +0.30 IQ points. These shifts further varied across countries and were systematically related to national-level macro indicators, such as socioeconomic (in-)equality, education, and healthcare access. The direction and strength of these associations differed across CHC stratum I subdomains. We further show substantial influences of measurement non-invariance and disentangle genuine cognitive change from artifacts of evolving test constructs and psychometric models. In all, we provide the first systematic evidence that the Flynn effect is domain-specific, while limited measurement invariance restricts meaningful interpretation of cross-temporal (domain-specific) trends.