Abstract The Earth's mantle contains significant amounts of carbon and is at depths greater than ∼250–300 km potentially so reducing that the Fe–C redox couple determines the nature of the reduced phase(s), which may be diamond, metal and carbides. Carbides will be Fe-rich but their stability also depends on the presence of Ni. We thus have experimentally investigated the Fe–Ni–C subsolidus ternary at 10 GPa, and secondly determined eutectic melting temperatures in this system. At subsolidus, the Fe-rich side of the ternary has two of the phases: diamond, Fe7C3 (to a molar XNi=Ni/(Fe+Ni)=0.11), Fe3C (to XNi=0.24) and metal stable, depending on bulk C-contents. At higher Ni-contents, (Fe, Ni)3C coexists with diamond and metal while at XNi⩾0.53, carbides are absent and diamond coexists with metal. Because Ni is more noble than Fe, it partitions strongly into the reduced phases such that at low metal fractions the metal phase reaches XNi>0.5 (at a bulk Ni-content of 1800 ppm for the mantle). Thermodynamic calculations at subsolidus conditions suggest that the mantle contains 50–700 ppm Fe, Ni metal at ∼300 km depth. Adopting bulk C contents of 50 to 500 ppm in the mantle would result in the phase association (Fe, Ni)3C + metal + diamond (at 10 GPa). An unexpected finding of this study is that eutectic temperatures in the Fe–Ni–C system are very low, 1210 °C at the Fe–C side, decreasing to 1125 °C at an XNi of 0.5 in the reduced phase. Hence we postulate that most of the deep reducing mantle will contain a small Fe–Ni–C melt fraction. These melts should be ubiquitous in the mantle, only those mantle regions where C-contents are less than what can be dissolved in the solid metal (50 ppm at 400 km depth) would not contain such a melt phase. However, the presence of a metal–carbon melt phase is probably of little long term consequence to mantle geochemistry as this melt is expected to remain in isolated pockets.