Abstract The solar system’s terrestrial planets are thought to have accreted over millions of years out of a sea of smaller embryos and planetesimals. Because it is impossible to know the surface density profile for solids and size frequency distribution in the primordial solar nebula, distinguishing between the various proposed evolutionary schemes has historically been difficult. Nearly all previous simulations of terrestrial planet formation assume that Moon- to Mars-massed embryos formed throughout the inner solar system during the primordial gas-disk phase. However, validating this assumption through models of embryo accretion is computationally challenging because of the large number of bodies required. Here, we reevaluate this problem with GPU-accelerated, direct N-body simulations of embryo growth starting from r ∼ 100 km planetesimals. We find that embryos emerging from the primordial gas phase at a given radial distance already have masses similar to the largest objects at the same semimajor axis in the modern solar system. Thus, Earth and Venus attain ∼50% of their modern mass, Mars-massed embryos form in the Mars region, and Ceres-massed objects are prevalent throughout asteroid belt. Consistent with other recent work, our new initial conditions for terrestrial accretion models produce markedly improved solar system analogs when evolved through the giant impact phase of planet formation. However, we still conclude that an additional dynamical mechanism such as giant planet migration is required to prevent Earth-massed Mars analogs from growing.