The Stale service has been launched for monitoring the relevance of AI model knowledge

Developers presented Stale — a specialized service for tracking release dates and temporal boundaries of training datasets for 20 popular AI models. The tool allows real-time matching of the model's release date with the moment of completion of data collection for its training. Analysis of the data shows a significant gap between the relevance of knowledge and the product release date. For example, the GPT-6 Astra model, presented in September 2026, has a "knowledge cutoff" in April 2026. This means that even the freshest systems can possess outdated information about events that occurred a few months before their official release. The service provides a visual table where exact temporal markers are indicated for each model. Users can use this data to assess the suitability of specific LLMs when working with up-to-date news summaries or rapidly changing technical specifications. The platform is available for public viewing and is regularly updated as new versions of neural networks are released. The Stale project helps developers and researchers better understand the limitations of models in tasks that require knowledge of the current time. The use of such tools allows avoiding hallucinations that occur when a model attempts to interpret events that occurred after its training was completed.