Performance of the random phase approximation (RPA) is tested for thermochemistry and geometries of transition-metal chemistry using various benchmarks obtained either computationally or experimentally. Comparison is made to popular (semi)local meta- and hybrid density functionals as well as to the second-order Møller-Plesset perturbation theory (MP2) and its spin-component-scaled derivatives. The benchmark sets include reaction energies, barrier heights, and dissociation energies of prototype bond-activation reactions, dissociation energies for a set of large transition-metal complexes, bond lengths and dissociation energies of metal hydride ions, and bond lengths and angles of a set of closed-shell first-row transition-metal complexes. The emphasis is on first-row transition-metal chemistry, though for energies, elements beyond the first-row are included. Attention is paid to the basis set convergence of RPA. For thermochemistry, RPA performs on par or better than the density functional theory (DFT) functionals presented and is significantly more accurate than MP2. The largest errors are observed in dissociation energies where the electronic environment is altered substantially. For structural parameters, very good results were obtained, and RPA meets the high quality of structures from DFT. In most cases, well-converged structures are obtained with basis sets of triple-zeta quality. MP2 optimized values can often not be obtained and are on average of inferior quality. Though chemical accuracy is not reached, the RPA method is a step forward toward a systematic, parameter-free, all-round method to describe transition-metal chemistry.