Today at #NeurIPS2025, we announced Titans, a brand-new architecture that combines the speed of RNNs with the performance of Transformers. It leverages deep neural memory for real-time learning and can efficiently scale to contexts of over 2 million tokens. For more information, please visit: