Abstract Turbulent heat transfer is a complex phenomenon, which is ubiquitous in engineering applications and has challenged turbulence modellers for several decades. In an attempt to simplify the problem it is often assumed that turbulent heat transfer can be inferred from the knowledge of the turbulent momentum transport, in what is known as the Reynolds analogy. This approach presents well-known drawbacks that limit its applicability to low-Prandtl fluids such as liquid metals. In an effort to overcome such limitations, an implicit Algebraic Heat Flux Model named AHFM-NRG has been recently proposed by the Nuclear Research and Consultancy Group (NRG). In the framework of the EU THINS project, this model was initially tested for a limited number of academic test cases in all three flow regimes (i.e.: natural, mixed and forced convection) and showed encouraging results. Further assessment and development of this or any other turbulent heat flux model with low-Prandtl fluids was hampered by the lack of accurate and reliable reference data. Thanks to the extensive reference database generated within the subsequent EU SESAME and MYRTE projects, the AHFM-NRG formulation has been further tested and developed. This article reports the development of the AHFM-NRG approach and its assessment against some representative test cases in all three flow regimes. It is shown that the AHFM-NRG formulations result in significant improvement with respect to the classical Reynolds analogy in all considered flow configurations.