Google introduces 'sleep framework' to help AI models retain knowledge continuously

On October 30, 2026, Google Research announced a new framework called 'Sleep Framework' designed to help large language models (LLMs) retain knowledge over time. The framework, introduced in a blog post on October 27, 2026, aims to address the issue of knowledge stagnation in AI models after deployment. Current models often face a 'knowledge cutoff' date, making it difficult to absorb new information without losing previously learned data. The solution draws inspiration from human memory consolidation during sleep, dividing model operation into 'awake' and 'sleep' phases. During the awake phase, models process input and store information temporarily, while the sleep phase allows for memory integration and self-improvement. The framework uses techniques like parameter expansion and knowledge distillation to preserve existing knowledge while integrating new data. Experiments showed the model can handle up to 10 million tokens of context and maintain near-zero performance drop in continuous learning tasks. The research was published under the title 'Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories'.