Native GPU-Accelerated Simulation and Estimation of Network Models R package cusna version 0.1.0 A self-contained native engine (a C interface over 'CUDA' kernels and C++ host logic) for stochastic actor-oriented models (the model family of 'RSiena'), exponential random graph models (cross-sectional, temporal, and separable temporal), and models for binary actor attributes, callable from R without a Python runtime. Modelled on the 'torch' package: the CRAN build is CPU-only from source; the GPU path is compiled from source when a 'CUDA' toolkit is detected at configure time. The data preparation, host statistics ('RSiena' Appendix B conventions), and moment targets are validated bit-for-bit against the reference implementation and reproduce 'RSiena' targets on public datasets to machine precision; the estimators match 'RSiena', 'ergm', 'btergm', and 'tergm' on public benchmark models. Software Imports: stats, utils LinkingTo: cpp11 Suggests: testthat (>= 3.0.0), jsonlite, litedown, RSiena (>= 1.4) Artem Maltsev <MaltsevSNA@proton.me> Comprehensive R Archive Network (CRAN) Artem Maltsev [aut, cre] MIT + file LICENSE (https://CRAN.R-project.org/package=cusna/LICENSE) 2026-07-15 application/tgz https://CRAN.R-project.org/package=cusna doi:10.32614/CRAN.package.cusna