What "How Much It Can Read at Once" Means
Context length is the upper limit on the text an LLM can refer to in a single request. According to Anthropic, the instructions, conversation history, and attached documents all fit in this space, along with the answer the model writes in that turn. It is separate from the vast text used in training; it is closer to the amount the model can keep spread out in front of it while working.
Length is stated in tokens. As of October 2026, the official documentation from Anthropic and Google lists models that can handle one million tokens. By this site's rule of thumb, that equals seven to ten Japanese paperback books.
Longer Is Not Always Better
Being able to accept a long document is not the same as reading it correctly. In a study by Liu et al. (published in 2023), performance was high when the needed information sat at the beginning or end of a document and dropped sharply when it sat in the middle. The pattern held even for models built to accept long inputs.
Anthropic also explains that accuracy and recall fall as the model is given more tokens to read. Choosing what to give the model is as important as the size of the space.
The "How Much It Can Read at Once" Condition on Our Pages
On the pages for choosing by use, "how much it can read at once" shows a model's context length in units of ten thousand tokens. When you set a minimum in the conditions, models shorter than that drop out of the table. For the use of having a model read documents and answer questions, the default narrows the list to 128,000 tokens or more. Values marked "about" are lengths rounded because they do not divide evenly into units of ten thousand tokens. Use length as a condition for whether a model has enough room, and compare reading ability itself on ratings that measure document comprehension.