This study presents a comparative evaluation of two transformer-based models, Legal-LED and Legal-Pegasus, for abstractive summarization of legal case documents. Using two domain-specific datasets: IN-ABS (Indian case law) and UK-ABS (UK judgments) we assess model performance in both pretrained and fine-tuned states. Results indicate that fine-tuned Legal-LED performs overall better on the IN-ABS dataset achieving a ROUGE-1 score of 35.39, ROUGE-2 of 16.85, ROUGE-L score of 21.06, ROUGE-LSUM of 33.25, demonstrating its strength in handling lengthy and procedurally complex texts. Legal Pegasus, while more effective in its pretrained form, shows consistent results on IN-ABS, particularly for shorter and more structured legal content. The analysis highlights the importance of model input capacity and fine-tuning for domain adaptation, offering insights into selecting appropriate summarization models for legal document processing.