Definition
Knowledge cutoff is a knowledge cutoff is the date after which an AI model's training data stops, so it knows nothing about later events unless it is given the information or can search the web.
A knowledge cutoff is the point in time where a model's training data ends. Anything that happened after that date is missing from what the model learned during training.
During pretraining, a model learns from a huge collection of text, code and other data gathered up to a certain date. Once training ends, its knowledge is frozen. Think of a printed encyclopedia: it is accurate as of the print date and silent about everything after.
Labs often publish a cutoff for each model, and it is usually a month or season rather than an exact day. The cutoff is not the same as the release date. Models typically launch months after their data ends, because training, post-training and safety testing take time. By the time you use a model, its knowledge may already be a year or more out of date.
A knowledge cutoff is about what the model learned. Fine-tuning can add some newer material, but it is not a reliable way to keep a model current. Retrieval and search are usually better, since they pull in fresh sources at question time rather than changing the model itself.
The cutoff is when the training data ends. The release date is when the model launches, usually months later.
Yes, if you paste in the information or the assistant can search the web or use live tools. Without those, it may not know or may guess.
Microsoft AI finished pre-training MAI-Base-2, a mixture-of-experts model with 2.8 trillion total and 80 billion active parameters, and says it is 4 to 12 times more compute-efficient than DeepSeek's base model. Nothing ships yet. The signal is that Microsoft wants its own frontier model.