A project named SETH (SESAR Thermal Hydraulics (SESAR: Senior Group of Experts on Nuclear Safety Research)) has been performed under the auspices of 15 OECD countries, with the aim of creating an experimental database suitable to assess the 3-D code capabilities in analyzing physical key phenomena relevant for LWR safety analysis. This paper presents the results of OECD/SETH Project Test 25 performed in the PANDA facility (located at PSI in Switzerland). Test 25 consists of two phases for duration of approximately 2 hours for each phase. During Phase 1 a steam-helium mixture (helium is used to simulate hydrogen) is released in a containment compartment initially filled with only air. This Phase simulate the hydrogen released in a postulated accident due to the fuel cladding metal-water (MW) reaction. During Phase 2 only steam is released in the containment and the scenario that occurs after the MW reaction is exhausted is addressed. Temperature and gas-concentration measurements obtained at several locations in the containment compartments allow the recognition of the complex stratification pattern evolution during the test period. The analysis of Test 25 carried out with the GOTHIC code is also presented in this paper. The analysis shows that the prediction of three-gas (hydrogen, steam, air) stratification pattern in a multi-compartment geometry for scenarios characterized by the evolution of the density difference between and inside compartments and in particular of the hydrogen accumulation in a dead-end volume is a challenging task also for a code having 3D capabilities.