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
Next step How to write code with AI → Work around stale knowledge: prefer stable libraries, and paste recent docs into the prompt.
Related terms
← All terms in the dictionary