The article evaluates the performance of AlphaFold 3, the latest version of AlphaFold, in predicting the 3D structures of RNA molecules. Despite AlphaFold's success in protein structure prediction, RNA structure prediction presents unique challenges due to fundamental differences between proteins and RNAs. The authors conduct a comprehensive benchmark using four different test sets and compare AlphaFold 3 with ten existing state-of-the-art methods, including ab initio, template-based, and deep learning approaches. The results show that while AlphaFold 3 outperforms most existing methods, it does not surpass human-aided methods in all metrics. AlphaFold 3 struggles with complex RNA interactions, particularly non-Watson-Crick interactions, and has difficulty predicting structures for long sequences and orphan structures. The study highlights the limitations of AlphaFold 3 and suggests areas for improvement in RNA structure prediction. The findings are available on the EvryRNA platform for further analysis.The article evaluates the performance of AlphaFold 3, the latest version of AlphaFold, in predicting the 3D structures of RNA molecules. Despite AlphaFold's success in protein structure prediction, RNA structure prediction presents unique challenges due to fundamental differences between proteins and RNAs. The authors conduct a comprehensive benchmark using four different test sets and compare AlphaFold 3 with ten existing state-of-the-art methods, including ab initio, template-based, and deep learning approaches. The results show that while AlphaFold 3 outperforms most existing methods, it does not surpass human-aided methods in all metrics. AlphaFold 3 struggles with complex RNA interactions, particularly non-Watson-Crick interactions, and has difficulty predicting structures for long sequences and orphan structures. The study highlights the limitations of AlphaFold 3 and suggests areas for improvement in RNA structure prediction. The findings are available on the EvryRNA platform for further analysis.