Knowledge cutoff
A knowledge cutoff is the point in time where a model's training data ends · the model cannot know about anything that happened later unless you give it that information.
Text reviewed October 5, 2026
A knowledge cutoff is the point in time where a model's training data ends · the model cannot know about anything that happened later unless you give it that information.
Basic
Language models learn from data collected up to a certain date. Ask about later events and the model either says it does not know or, worse, guesses. Labs usually publish a cutoff date, but the date a model really knows about can be earlier, because recent months are thinly covered on the web when the data is collected.
Deep
The stated cutoff and the effective cutoff can differ. BenchGecko measures the effective one in the Knowledge Horizon test: each model is asked about dated news events month by month, and the test finds where its answers stop being right. Retrieval (RAG) and web search tools are the usual way to give a model information newer than its cutoff.
Expert
Effective knowledge decays gradually before the cutoff because the volume of text about an event keeps growing for months after it happens. A model trained soon after an event has seen less about it than a model trained a year later.
Depending on why you're here
- ·The last date the AI learned about
- ·It does not know newer news unless you tell it
- ·Use retrieval or web search for anything recent
- ·Check the measured cutoff on /gecko-tests/knowledge-horizon
- ·Older cutoffs make a model less useful for news and markets
- ·Stated vs effective cutoff can differ
- ·Measured per model by the Knowledge Horizon test