A new framework, BabelCoder, has been introduced for code translation using large language models. It decomposes the task into specialized agents for translation, testing, and refinement, improving translation quality. BabelCoder outperforms existing methods in 94% of cases with an average accuracy of 94.16%.