Google advertises a 1M-token context window for Gemini 3.1 Pro, but uploading a 100-page PDF on gemini.google.com can return Files and prompt exceed Gemini's context window. For better results, try uploading smaller files. That is not false advertising. The 1M window in the consumer Gemini app is gated to paid tiers, and free accounts get a much smaller window.
As of June 2026 Google publishes the per-tier numbers in the Gemini Apps Help Center: Free = 32K tokens, Google AI Plus = 128K tokens, Google AI Pro and AI Ultra = the full 1M tokens (about 1,500 pages of text or 30,000 lines of code). The fastest fix is to know which bucket you are in and route the task to a surface that has enough room.
Fastest fix: if you are on Free, do your long-document work in aistudio.google.com (the full window, no subscription) or upgrade to Google AI Pro ($19.99/month) for the 1M window inside the app.
Which bucket are you in
Match your symptom to the right cause, then jump to the matching step:
| If this is true | Most likely cause | Go to |
|---|---|---|
| Top model picker shows Flash or a non-Pro model | Smaller model and/or smaller window | Step 1 |
| Account page shows “Free”, no subscription | 32K free-tier cap | Step 2 |
| Account shows “Google AI Plus” | 128K cap (not the full 1M) | Step 2 |
| The PDF is scanned (no selectable text) | Image tokens, ~1-2K per page | Step 4 |
| Chat already has many turns or several uploads | History ate the window | Step 7 |
| Work/Workspace account only | IT-set cap | Step 8 |
| You need long context routinely | Wrong surface; use API / AI Studio | Step 6 |
Common causes
By frequency:
1. You are not on a Pro/Ultra plan (most common)
This is the single biggest factor as of June 2026. Google now publishes the per-tier window in the Gemini app:
| Gemini app tier | Context window | Roughly |
|---|---|---|
| Free | 32K tokens | ~50 pages of text |
| Google AI Plus | 128K tokens | ~190 pages |
| Google AI Pro | 1M tokens | ~1,500 pages |
| Google AI Ultra | 1M tokens | ~1,500 pages |
The full 1M number you see in marketing is the Pro/Ultra number. Free is roughly 30x smaller. The Web UI does not tell you the window is small; it just refuses the file.
How to judge: open gemini.google.com/app, check the plan badge under your account avatar, or visit one.google.com/about/google-ai-plans.
2. Wrong model selected (smaller or compressed window)
If the top model picker is set to a Flash / fast model rather than Gemini 3.1 Pro, you can get a smaller effective window and weaker long-document handling, even on a paid plan.
How to judge: top model picker shows something other than the Pro / Thinking model.
3. Attachments cost tokens by parsed content, not display size
The file-size limit and the token cost are two different things. A PDF can be small on disk but expensive in tokens once parsed.
- Display size: 5 MB
- Actual context cost: parsed text plus any image/OCR tokens, sometimes 200K+
Rough token cost for a 100-page PDF: 30K-60K tokens text-only; 80K-200K tokens for mixed text and images. Either of those overflows a 32K free window immediately.
How to judge: upload returns Files and prompt exceed Gemini's context window even though the file is well under the 100 MB per-file size limit.
4. Scanned PDFs / image-heavy docs
Scanned PDFs have no text layer, so Gemini treats each page as an image at roughly 1-2K tokens per page (image encoding). 100 scanned pages can be 100-200K tokens, which overflows Free (32K) and Plus (128K).
How to judge: in the PDF you cannot select or search the text (it is a picture of text).
5. Conversation history filled the window
The Gemini app counts every prior turn in the current conversation against your window. If you already uploaded a few PDFs in this chat, a new file fights for whatever is left.
How to judge: the current conversation has many turns or several earlier file uploads.
6. Workspace-managed accounts can have stricter caps
A corporate Workspace can set a lower context or file cap for data-exfiltration reasons.
How to judge: the same file uploads fine on your personal account but fails on the work account. Confirm with IT.
Shortest path to fix
Ordered by context size unlocked, cheapest first.
Step 1: Select Gemini 3.1 Pro
gemini.google.com → top model picker → Gemini 3.1 Pro (Thinking)Pro is the model that actually reaches the 1M window on paid tiers. Flash / fast models trade window and depth for speed.
Step 2: Upgrade your plan to unlock the window
one.google.com/about/google-ai-plans → choose a planAs of June 2026:
- Google AI Plus (
$4.99/month after June 8, 2026): raises the app window to 128K tokens (~190 pages). Enough for a single medium report, not a 100-page PDF. - Google AI Pro (
$19.99/month, formerly “Gemini Advanced” / “Google One AI Premium”): unlocks the full 1M-token window (~1,500 pages) plus Gemini 3.1 Pro. - Google AI Ultra (
$99.99/month): same 1M window, higher usage limits and extra features.
If your goal is one large document, Pro is the smallest plan that gives the full window. Plus at 128K will still reject a 100-page PDF.
Step 3: Split large documents (works on any tier)
Break a 100-page PDF into batches small enough for your window:
# pdftk (split into two 50-page batches)
pdftk input.pdf cat 1-50 output batch1.pdf
pdftk input.pdf cat 51-100 output batch2.pdf
# Or macOS Preview / Adobe Acrobat / ilovepdf.comWorkflow:
- Upload
batch1.pdfand ask: “summarize this section, output a 1K-word brief” - Copy the brief
- New conversation: upload
batch2.pdfplus the brief, ask Gemini to merge - Repeat across all batches
On a 32K free window, use smaller batches (10-20 pages).
Step 4: OCR scanned PDFs before uploading
# ocrmypdf (open source)
ocrmypdf input.pdf output_ocr.pdf
# Or Adobe Acrobat → Scan & OCR → Recognize TextAfter OCR the text layer is read as plain text, dropping a scanned 100-pager from ~200K image tokens to roughly 30-60K text tokens.
Step 5: Convert PDF to plain text / markdown
# pdftotext
pdftotext input.pdf output.txt
# Then upload the .txtPlain text has the lowest token overhead. About 1 MB of plain text is roughly 250K tokens, which fits the 1M Pro/Ultra window with room to spare but still overflows Free and Plus, so combine with Step 3 if needed.
Step 6: Use Google AI Studio or the Gemini API (full window, free)
This is the best fix for routine long-context work, and it costs nothing for the UI:
- Go to aistudio.google.com, select Gemini 3.1 Pro
- Upload the 100-page PDF directly; the token counter shows your usage against the 1M window
- No subscription needed (note: since March 23, 2026 new AI Studio users may be prompted to set up a prepaid billing plan before first use)
For automation, call the API. As of June 2026 Gemini 3.1 Pro is $2/1M input tokens and $12/1M output tokens:
from google import genai
client = genai.Client(api_key="YOUR_KEY")
# Files API for large uploads
file = client.files.upload(path="huge_doc.pdf")
response = client.models.generate_content(
model="gemini-3.1-pro",
contents=[file, "Summarize key points"],
)
print(response.text)The API and AI Studio both give you the full 1M window regardless of which consumer plan you are on.
Step 7: Start a new conversation to free up history
If the current chat is bloated with old turns and files:
- Start a new conversation
- Summarize the prior context into a brief of at most 5K tokens
- Reuse that brief plus the new file in the fresh chat
Step 8: Workspace — ask IT to raise the cap
If only the work account fails, IT controls it: Admin Console → Apps → Google Workspace → Gemini app, where they can review file-size and context settings for your org.
How to confirm it is fixed
- The upload completes without the
Files and prompt exceed Gemini's context windowmessage. - Ask a needle-in-haystack question whose answer sits on a late page (for example, “what does page 92 conclude?”). A correct, specific answer means the whole document is in the window, not just the first chunk.
- In AI Studio, the token counter shows your input well under 1,048,576 tokens.
FAQ
Why does Gemini say 1M tokens but reject my 100-page PDF? The 1M window only applies to Google AI Pro and Ultra. On Free you get 32K tokens (about 50 pages) and on Google AI Plus 128K (about 190 pages), so a 100-page PDF overflows both. These per-tier numbers are published in the Gemini Apps Help Center as of June 2026.
Is there a free way to get the full 1M window? Yes. Use aistudio.google.com with Gemini 3.1 Pro. The UI is free and exposes the full 1M window, with a live token counter. New AI Studio users since March 23, 2026 may be asked to set up a prepaid billing plan first.
What is the exact file limit in the Gemini app? Up to 10 files per prompt, 100 MB each (2 GB for video). The size limit and the token limit are separate; a file can be under 100 MB but still overflow your token window.
Does Google AI Plus give me the long context window?
No. Plus is 128K tokens (~190 pages), not the full 1M. For a 100-page PDF or larger you need Google AI Pro ($19.99/month) or Ultra.
My PDF is small on disk but still rejected. Why? Token cost depends on parsed content, not file size. Scanned or image-heavy PDFs cost roughly 1-2K tokens per page as images. OCR the file (Step 4) or convert it to plain text (Step 5) to cut the cost.
Prevention
- Routine long-context work belongs in aistudio.google.com, not gemini.google.com. It is free and uses the full 1M window.
- OCR scanned PDFs first; it saves roughly 70% of the tokens.
- For text-only tasks, convert to
.txtrather than uploading a PDF. - If you need the 1M window inside the app, Google AI Pro (
$19.99/month) is the cheapest tier that unlocks it. Google AI Plus at 128K is not enough for a 100-page document. - Heavy long-doc work (papers, 10-Ks, codebases) belongs on the Gemini API for the full window and automation.
Related
Tags: #Gemini #Troubleshooting