Context : Luminous red novae ( LRNe ) have absolute magnitudes intermediate between novae and supernovae , and show a relatively homogeneous spectro-photometric evolution .
Although they were thought to derive from core instabilities in single stars , there is growing support to the idea that they are triggered by binary interaction , possibly ending with the merging of the two stars .

Aims : AT 2018hso is a new transient showing transitional properties between those of LRNe and the class of intermediate luminosity red transients ( ILRTs ) similar to SN 2008S .
Through the detailed analysis of the observed parameters , our study support that it actually belongs to the LRN class , and was likely produced by the coalescence of two massive stars .

Methods : We obtained ten months of optical and near infrared photometric monitoring , and eleven epochs of low-resolution optical spectroscopy of AT 2018hso .
We compared its observed properties with those of other ILRTs and LRNe .
We also inspected archive Hubble Space Telescope ( HST ) images obtained about 15 years ago to constrain the progenitor ’ s properties .

Results : The light curves of AT 2018hso show a first sharp peak ( M _ { r } = -13.93 mag ) , followed by a broader and shallower second peak , that resembles a plateau in the optical bands .
The spectra dramatically change with time .
Early time spectra show prominent Balmer emission lines and a weak [ Ca ii ] doublet , which is usually observed in ILRTs .
However , the major decrease in the continuum temperature , the appearance of narrow metal absorption lines , the major change in the H \alpha strength and profile , and the emergence of molecular bands support a LRN classification .
The possible detection of an I \sim - 8 mag source at the position of AT 2018hso in HST archive images is consistent with expectations for a pre-merger massive binary , similar to the precursor of the 2015 LRN in M101 .

Conclusions : We provide reasonable arguments to support a LRN classification for AT 2018hso .
This study reveals growing heterogeneity in the observables of LRNe than thought in the past , making sometimes tricky the discrimination between LRNe and ILRTs .
This suggests the need of monitoring the entire evolution of gap transients to avoid misclassifications .