What is a context window?
A context window is how much text an AI model can see and work with at once: your instructions, the conversation so far, any files, and its own reply. It's measured in tokens, small chunks of text of roughly four characters each.
Updated September 24, 2026
A context window is an AI model's short-term working memory: the total amount of text it can see and use at one time. Everything counts toward it, including the system instructions, every earlier message in the chat, documents or images you add, and the reply the model is writing.
Tokens: how it's measured
Models don't read words, they read tokens: small chunks of text that can be a single letter, part of a word or a whole short word. Google's rule of thumb is that one token is about four characters, and 100 tokens is roughly 60 to 80 English words. Token counts vary by language and by model.
How big are they?
As of September 2026, many leading models accept around a million tokens:
- Anthropic lists a 1 million token window for current Claude models such as Opus 5.5 and Sonnet 5.
- Google lists 1,048,576 input tokens for Gemini 3.8 Flash.
- OpenAI lists 1.05 million tokens for its GPT-6 models.
For scale, Google says a million tokens is about eight average-length novels or 50,000 lines of code.
Why bigger isn't always better
- Quality drops with length. Anthropic says accuracy and recall get worse as the token count grows, and IBM notes models do worst on details buried in the middle of a long input.
- Cost and speed. Longer inputs cost more and take longer before the first word appears.
- Hard limits. Go over the window and the request fails or the reply gets cut off. Chat apps may drop the oldest messages instead.
That's why tools like RAG exist: instead of pasting everything in, they find and send only the passages that matter.
Quick questions
How many words is 1,000 tokens?
About 600 to 800 English words, based on Google's estimate of 60 to 80 words per 100 tokens. Other languages can use more tokens for the same text.
Does a bigger context window mean a smarter model?
No. It lets the model take in more at once, but Anthropic warns that recall and accuracy tend to drop as inputs get very long.
What happens when you go over the context window?
Developer tools usually reject the request or cut off the reply. Chat apps often quietly drop the oldest messages, which is why a long chat can forget what you said at the start.
Is the context window what the model was trained on?
No. It's the working memory for one conversation, separate from the huge body of data the model learned from.
Sources: Anthropic: Context windows, Google: Understand and count tokens, Google: Long context, IBM: What is a context window?