In this chapter, the circuit predictions from optimized models will be evaluated using the test portion of the dataset and inference predictions, which were assessed using Cadence’s Spectre simulator. This simulator provides industrial-level accuracy for the performances, which are then compared to the target performance for final model evaluation. The models evaluated are the three diffusion models, Multi-Layer Perceptron (MLP), Residual Multilayer Perceptron (ResMLP) and Encoder-only Transformer (EoT), and the Supervised Learning (SL) models. It is important to note that since these circuits are composed solely of MOSFET devices operating in saturation, all tests began by filtering out any generated and predicted points considered outside the saturation margins before calculating the error. This filtering was necessary because the proposed model was trained exclusively on a dataset within the saturation zones. Points outside these zones could exhibit different circuit behavior, potentially skewing the results. This was accomplished by removing circuit sizing with values smaller than the saturation values for overdrive voltage and the saturation margin: 50mV for both values in the Voltage Combiners biased Operational Transconductance Amplifier (VCOTA) [1] and 50 and 80 mV for the folded architecture [2], respectively. This chapter it is divided into two sections. The first section focuses on the VCOTA, with the second part analyzing the Folded VCOTA results.