Context. Direct dynamical mass measurements of stars with masses above 30 M_⊙ are rare. This is the result of the low yield of the upper initial mass function and the limited number of such systems in eclipsing binaries. Long-period, double-lined spectroscopic binaries that are also resolved astrometrically offer an alternative to eclipsing binaries for obtaining absolute masses of stellar objects. 9 Sgr (HD 164794) is one such long-period, high-mass binary. Unfortunately, a large amount of tension exists between its total dynamical mass inferred spectroscopically from radial velocity measurements and that from astrometric data. Aims. Our goal is to resolve the mass tension of 9 Sgr that exists in literature, to characterize the fundamental parameters and surface abundances of both stars, and to determine the evolutionary status of the binary system, henceforth providing a reference calibration point to confront evolutionary models at high masses. Methods. We obtained the astrometric orbit from existing and new multi-epoch VLTI/PIONIER and VLTI/GRAVITY interferometric measurements. Using archival and new spectroscopy, we performed a grid-based spectral disentangling search to constrain the semi-amplitudes of the radial velocity curves. We computed atmospheric parameters and surface abundances by adjusting FASTWIND atmosphere models and we compared our results with evolutionary tracks computed with the Bonn Evolutionary Code (BEC). Results. Grid spectral disentangling of 9 Sgr supports the presence of a 53 M_⊙ primary and a 39 M_⊙ secondary, which is in excellent agreement with their observed spectral types. In combination with the size of the apparent orbit, this puts 9 Sgr at a distance of 1.31 ± 0.06 kpc. Our best-fit models reveal a large mass discrepancy between the dynamical and spectroscopic masses, which we attribute to artifacts from repeated spectral normalization before and after the disentangling process. Comparison with BEC evolutionary tracks shows the components of 9 Sgr are most likely coeval with an age of roughly 1 Myr. Conclusions. Our analysis clears up the contradiction between mass and orbital inclination estimates reported in previous studies. We detect the presence of significant CNO-processed material at the surface of the primary, suggesting enhanced internal mixing compared to currently implemented in the BEC models. The present measurements provide a high-quality high-mass anchor to validate stellar evolution models and to test the efficiency of internal mixing processes.