The chat forgets
Earlier turns fall out of the window. On Free that happens after roughly 24,000 words of conversation and pasted text combined.
Last verified September 1, 2026
Gemini context window by plan
Usage limits are multipliers, but the context window is a hard, published number: 32k tokens on Free, 128k on AI Plus, 1M on AI Pro and Ultra. It is the limit you hit on long documents, and no refresh clears it.
Context window
The limit Google does publish exactly: context per plan.
Usage limits are multipliers, but the context window is a hard number and it is the difference you feel first on long documents.
| Plan | Context window | What fits |
|---|---|---|
| Free | 32,000 tokens | About 24,000 words: a long article or a short report |
| Google AI Plus | 128,000 tokens | About 96,000 words: a book-length manuscript or a large codebase slice |
| Google AI Pro | 1,000,000 tokens | Roughly 750,000 words, hours of video or a whole repository |
| Google AI Ultra (both tiers) | 1,000,000 tokens | Same as Pro; Ultra buys usage, not context |
When a long chat “forgets” or a big upload fails on Free, that is the 32k window, not your usage allowance. Splitting the document or moving to Plus fixes it; waiting five hours does not.
Symptoms
How a context limit shows up.
An upload fails or is truncated
A 200-page PDF is far beyond 32k tokens. Plus (128k) handles a book; Pro and Ultra (1M) handle a repository or hours of video.
Answers ignore part of the document
The model only sees what fits. Ask about one section at a time on Free, or upgrade.
Waiting does nothing
This is not usage. The five-hour refresh and weekly cap are separate; the window is fixed per plan.
Fixes
Work inside the window, or buy a bigger one.
| Situation | Fix |
|---|---|
| Long article or report on Free | Paste one section at a time; ask for a running summary |
| Book-length manuscript | AI Plus at $7.99 gives 128k, enough for most manuscripts |
| Whole codebase, long video, many documents | AI Pro at $19.99 gives 1M; Ultra adds usage, not context |
| Multiple large files at once | Pro or Ultra, and a fresh chat per project so the window is not shared |
FAQ
Short answers
Does Ultra have a bigger context window than Pro?
No. Both are 1M tokens per Google’s help page; Ultra buys 5x or 20x the usage, not more context.
Is the context window the same as the usage limit?
No. Usage refreshes every five hours under a weekly cap; the context window is fixed per plan and never refreshes.
How many words is 32k tokens?
Roughly 24,000 English words, including both what you paste and the conversation so far.
Does Deep Research have its own window?
Deep Research reads sources outside the chat window, but its report and your follow-ups still live in the chat’s context.