Introduction: Syntax, the ability to combine discrete linguistic elements into hierarchically structured expressions, is a defining feature of human language. In cognitive neuroscience, syntactic processing encompasses the retrieval and manipulation of grammatical knowledge during language comprehension and production. Despite decades of research, the neural architecture underlying syntactic processing in the human brain remains unclear. Here, we sought to identify the neural basis and dynamics of syntactic processing by means of a pre-registered, large-scale activation likelihood estimation (ALE) meta-analysis with meta-analytic connectivity modeling (MACM). Methods: Using GingerALE (version 3.0.2; Eickhoff et al., 2009), we performed ALE and MACM on 101 PET and fMRI studies containing 104 experiments and 277 relevant contrasts, derived from the dataset reported in Turker et al. (2023, Psychol. Bull.). We used the recommended ALE voxel-level cluster forming threshold of p<0.001 and cluster-level threshold of p<0.05 FWE-corrected (Müller et al., 2018). We investigated both the functional architecture and connectivity profile of syntactic processing across three levels of syntactic specificity, stratified by how precisely a given contrast isolated combinatorial processes: low (contrasts with non-linguistic baselines); medium (contrasts with linguistic baselines, but matching low-level stimulus features, e.g. sentences > non-words); and high (contrasts which are explicitly combinatorial, or address a complex syntactic process, e.g. MOV > BIND). Specificity levels were defined and rated by four independent coders and achieved high interrater reliability (Gwet’s AC1 coefficient = 0.85). We assessed overlap between the core syntax network - regions engaged consistently across all specificity levels - and the domain-general multiple demand network (MDN) (map by Lipkin et al. 2022, Sci. Data; thresholded at 30% probability). Results: We identified an exclusively left-hemispheric core network for syntax engaged across all specificity levels, comprising the inferior frontal gyrus (IFG; BA 44, 45, and OP9) and sulcus (IFS), middle frontal gyrus (MFG) and precentral sulcus (PreCS), insula, pre-supplementary motor area (pre-SMA), and posterior superior temporal sulcus (pSTS). The core syntax network overlapped with the domain-general MDN in the left IFS, insula, and pre-SMA, but not in left IFG, MFG/PreCS, and pSTS. Meta-analytic contrasts showed that high syntactic specificity preferentially recruited sub-regions of left posterior IFG, pSTS, MFG/PreCS and intraparietal sulcus, compared to low and medium specificity contrasts. MACM characterized the connectivity profiles of the regions comprising the core network, providing converging evidence for co-activation of the core syntax nodes in left IFG, pSTS, MFG/PreCS, and pre-SMA during language processing. Discussion: Our findings reveal a left-lateralized core network for syntax, with domain-specific components in left IFG, MFG/PreCS and pSTS, and domain-general executive components in left IFS, insula, and pre-SMA. Meta-analytic connectivity modeling corroborates the coordinated recruitment of these regions as a coherent syntactic network during language processing. In sum, this work provides a comprehensive characterization of the neural architecture underlying human syntax and points towards a hybrid neural architecture of grammar in which domain-specific and domain-general networks work in tandem even during highly syntax-specific tasks.