Résumé
In this review presented at the Symposium A Stellar Journey in Uppsala, June 2008, I give an account of the historical development of the MARCS code, and its premises from the first version published in 1975 to the 2008 grid. The primary driver for the development team who constantly strive to include the best possible physical data, is the science that can be done with the models. A few preliminary comparisons of M star model spectra to spectrophotometric observations are presented. Particular results related to opacity effects are discussed. The size of errors in spectral energy distribution (SED) and model thermal stratification is estimated for different densities of wavelength sampling. The number of points used in the MARCS 2008 grid (108 000) is large enough to ensure errors of only a few K in all models of the grid, except the optically very thin layers of metal-poor stars. Errors in SEDs may reach about 10% locally in the UV. The published sampled SEDs are thus adequate to compute synthetic broadband photometry, but higher resolution spectra will be computed in the near future and published as well on the MARCS site (marcs.astro.uu.se). Test model calculations with TiO line opacity accounted for in scattering show significant cooling of the upper atmospheric layers of red giants. Rough estimates of radiative and collisional time scales for electronic transitions of TiO indicate that scattering may well be the dominant mechanism in these lines. However, models constructed with this hypothesis are incompatible with optical observations of TiO (Arcturus) or IR observations of OH (Betelgeuse), although they may succeed in explaining H2O line observations. More work is needed in that direction.