Quick definition · 2 min AI term
Training cutoff
A model’s training cutoff is the date its training data stops, so it knows nothing that happened after it.
Think of it like
Like a textbook printed on a certain date: sharp on everything up to then, blank on anything newer.
Example
Ask a model with an early-2024 cutoff about a library released in 2025 and it will guess, often inventing functions that do not exist.
Why it matters
It explains a lot of confident-but-wrong coding answers: the model is recalling an older version of the world. Knowing the cutoff tells you when to paste in fresh docs.
Where you’ll see it
ChatGPTClaudeCursormodel release notes
Related terms