This study presents the first comprehensive description of the operational GEMS (Geostationary Environment Monitoring Spectrometer) ozone profile retrieval algorithm and evaluates the performance of the recently reprocessed version 3.0 dataset. The retrieval operates in the 310–330 nm spectral range and yields total degrees of freedom for ozone ranging from 1.5 to 3. Although the vertical sensitivity is limited, GEMS achieves an effective vertical resolution of 5–10 km and is capable of separating tropospheric and stratospheric ozone layers. This work primarily highlights the substantial algorithmic and calibration enhancements introduced in version 3.0 over the previous version, including improvements to the slit function, wavelength calibration, and radiometric calibration. In particular, the irradiance offset has been a major issue affecting the accuracy of ozone profile and other Level 2 products. To address this, the measured irradiance is scaled relative to a high-resolution solar reference spectrum using a correction factor. Residual wavelength-dependent biases in the normalized radiance are further addressed through soft calibration. As a result, version 3.0 significantly reduces spectral fitting residuals, lowering them from 0.8 % in version 2.0 to 0.2 % under nominal conditions. This improvement also mitigates the altitude-dependent oscillating biases observed in the previous version, which included up to 40 DU overestimation in the troposphere and 20 DU underestimation in the stratosphere, when compared with ozonesonde observations. The version 3 ozone profiles show agreement within 10 DU of ozonesonde profiles, with a mean bias of −7.7 % in tropospheric ozone columns and within 5 % in the stratosphere. Furthermore, the retrievals capture day-to-day vertical ozone variability, as demonstrated by comparisons with daily ozonesonde launches in February and March 2024. Integrated ozone columns derived from the profiles also show improved consistency with ground-based total ozone measurements, yielding a mean bias of −3.6 DU and outperforming the GEMS operational total column ozone product.