According to the World Health Organization, 99% of the global population lives in areas where air pollution levels exceed WHO guidelines. Air quality has become a critical global concern, with pollutants such as PM2.5, PM10, SO2, CO, NO, and Ozone contributing to severe health conditions like heart disease and respiratory illnesses. The Air Quality Index (AQI) is an essential tool for the public to understand current pollution levels and take preventive measures. This paper focuses on forecasting AQI for Delhi’s ITO region using deep learning methods, including Long Short-Term Memory (LSTM) networks, and evaluating language models (LMs) such as LagLlama, Gemma, and Llama 3.1 for time-series forecasting leveraging zero shot prompting, fine-tuning and in context learning. The dataset spans a 5 year period from January 1, 2019, to December 31, 2023. Results demonstrate that LLM-based models, particularly Llama 3.1, yield lower prediction errors. However, improvements are needed in capturing long-term pollution trends. Fine-tuned LagLlama consistently outperforms other methods, offering reduced errors in forecasting pollutant levels. While LSTM models slightly edge out LagLlama in performance, the fine-tuned LagLlama model enables users to anticipate peak pollution hours and plan their activities during periods of low air quality. This tool provides valuable insights into upcoming air quality trends, serving as a helpful resource for policymakers and the general public in managing air quality.