Knowledge cutoff
The knowledge cutoff is the date after which a language model has no training data — anything that happened after is unknown to it unless supplied at inference time.
Every base LLM has a knowledge cutoff: Claude 4.6 ~early 2025, GPT-5 ~mid-2025, Gemini 2 ~late 2024. Asking about events after the cutoff produces confident hallucination unless the model has retrieval / web search. Production apps that handle current information must either route to a model with web access (Claude with web, Gemini with AI Overviews, Perplexity), implement RAG over fresh sources, or refuse questions outside the cutoff. The cutoff is one of the few model facts every developer should know off the top of their head for the models they use.
Common mistakes
- Assuming a model knows recent news.
- Forgetting to update RAG sources when the model cutoff lags.
FAQ
What is knowledge cutoff?
The knowledge cutoff is the date after which a language model has no training data — anything that happened after is unknown to it unless supplied at inference time.
What are the most common mistakes with knowledge cutoff?
Assuming a model knows recent news. Forgetting to update RAG sources when the model cutoff lags.
Related terms
- Retrieval-augmented generation (RAG) — Retrieval-augmented generation (RAG) injects relevant documents into the prompt at query time so the model answers from your data instead of its training memory.
- Hallucination — A hallucination is when a language model produces output that is factually wrong, fabricated, or unsupported, while sounding confident.
- Grounding — Grounding is any technique that ties a language model's output to verifiable sources — retrieved documents, tool results, structured data — instead of pure memory.
Last updated: 2026-06-01. Raw markdown: https://promtable.com/glossary/knowledge-cutoff.md.