The European Space Agency (ESA) Sentinel-5 Precursor (S5P) is a low Earth orbit polar satellite carrying the single payload instrument TROPOspheric Monitoring Instrument (TROPOMI). Since its launch on 13 October 2017, the S5P mission has been acquiring almost 8 years of nadir ozone profile data, retrieved from the UV bands 1–2 measurements in the spectral range 270–330 nm. The retrieval algorithm of the ozone profile is strongly affected by systematic effects in the measured radiance, therefore absolute calibration of the input spectra is necessary to obtain good quality retrievals. In this study, we characterize the radiometric bias of the TROPOMI bands 1–2 measurements in comparison with simulations obtained with the Determining Instrument Specifications and Analysing Methods for Atmospheric Retrieval (DISAMAR) radiative transfer model. This comparison is the basis of the so-called “soft” calibration correction, an empirical correction applied at level-2 (L2), before the retrieval. The soft calibration correction reduces the reflectance fit residuals of 20 %–30 %, which improves the precision of the integrated total and tropospheric ozone columns of 10 %–15 %, as well as reducing the along-track orbit artifacts. The soft calibration correction spectra provide useful insights into the instrument radiometric calibration and can be used together with the in-flight calibration measurements to investigate and enhance the radiometric calibration, especially in band 1 where it shows a large spectral, radiance, across-track position and temporal dependence. From the comparison between on-ground and in-flight calibration measurements, some inconsistencies were found in the L1 calibration of the bands 1–2 which were traced to the straylight and the residual signal correction algorithms and are the subject of this study. Bands 1–2 measurements have been reprocessed with improved L1 correction algorithms to address the remaining uncorrected additive effects. The soft calibration correction spectra, derived from this reprocessed L1 data, are significantly reduced in magnitude (around 15 %–20 %, especially in band 1), and show less across-track position and spectral/temporal biases. Even if the soft calibration is still an essential pre-processing step for the ozone profile retrieval algorithm, the retrieval obtained with the updated version shows decreased dependence on the correction and an overall enhancement of the global retrieval convergence. The updates to the L1 and soft calibration are included in the ESA's official upgrade of the L1b processsor version 3.0 and ozone profile algorithm processor version 2.9.0, which will be also used for the second TROPOMI mission reprocessing.