Worldwide health problems and feelings of worry and anxiety have been brought on by COVID-19, a terrible pandemic that the WHO has declared. Sentiment analysis is a vital technique for figuring out how people are responding to the pandemic. In order to find the most pertinent and instructive aspects of the embeddings, the research suggests a novel sentiment analysis model for COVID-19 tweets using CT-BERT as a base model and MAX Pooling function on the last four layers. The generated embeddings are joined with the classification (CLS) token before being sent to a classifier, which generates a probability distribution across all potential classes. The proposed technique acquired 91 and 92 % accuracy, 93 and 94% recall and 90 and 92% F-measure for positive and negative sentiment classification respectively. Insights into public sentiment and emotions have been gained during COVID-19 that have been useful for informing decision-making and communication tactics.