Try Kimi at kimi.com →

How to Analyze PDFs
With Kimi AI

Upload a report, paper or contract and let Kimi summarise it, answer questions, pull out tables and compare versions, with page numbers you can check. Here's how, with file limits, ready-to-copy prompts and fixes for common problems.

Up to 100 MB per file Up to 50 files Free to start
PDF annual-report.pdf · 120 pages read KKimi reading 120 pages… SUMMARY KEY NUMBERS Revenue$4.2Bp.12 Net profit$610Mp.14 Staff9,800p.31 What are the main risks? p.40 ✓ Every answer points to a page you can check
STEPS

How to analyze a PDF with Kimi AI

  1. 1

    Open Kimi and sign in

    Go to kimi.com or open the Kimi app and sign in. A free account is enough to start.

  2. 2

    Pick a model

    Kimi K3.1 Pro is the default and handles long documents well. For very long files, Max and Ultra plans get the full 1M-token chat context.

  3. 3

    Attach your PDF

    Click the + button or drag the file into the chat box. Each file can be up to 100 MB, and you can add up to 50 files.

  4. 4

    Ask a specific question

    Say what you want and in what format, for example “List the 5 main risks with page numbers.” Vague requests get vague answers.

  5. 5

    Check and follow up

    Open the pages Kimi points to and confirm the key facts. Then ask follow-up questions in the same chat.

A good first prompt

Using only the attached PDF, give me a 5-sentence summary, the 7 key findings with page numbers, and anything that looks unusual or risky. Tell me if something isn't covered.

New to Kimi? Start with the beginner's guide, or see what the free plan includes.

LIMITS

Kimi PDF and file limits

File size is rarely the problem. For very long PDFs, what matters is how much text the model can hold at once.

LimitWhat it isSource
File typesPDF, Word, Excel, PowerPoint, images, TXT and videoKimi Help Center
File sizeUp to 100 MB per fileKimi Help Center
Files at onceUp to 50 filesKimi Help Center
Page limitNot publishedVery long files depend on the model's context, not a page count
Chat contextUp to 1M tokens on Max and Ultra plans; less on Free, Plus and ProKimi pricing page
ProjectsFree: 2 projects, 500 MB. Paid plans: up to 100 projects and 50 GBKimi Help Center
API (Files API)100 MB per file, 10 GiB storage per organization by defaultKimi API docs

Plan details are on the Kimi pricing guide. A 1M-token context fits a very long document, but Kimi doesn't publish a page count because pages vary so much in how much text they hold.

PROMPTS

9 prompts for analyzing PDFs

Copy a prompt, attach your PDF and replace the [brackets]. Each card shows where to run it in Kimi.

Summary

Summarise a long report

Use in: Chat

Summarise the attached PDF for a busy manager: a 5-sentence overview, the 7 key findings with page numbers, and 3 questions I should ask the authors.

Tables

Pull tables into a spreadsheet

Use in: Agent · Sheets

Find every table in the attached PDF about [topic]. Rebuild them as one spreadsheet with clear column names, add the source page for each row, and flag any cell you couldn't read clearly.

Q&A

Ask questions with page proof

Use in: Chat

Using only the attached PDF, answer: [your question]. Quote the exact sentence you relied on and give its page number. If the PDF doesn't say, tell me.

Compare

Compare two versions

Use in: Chat

Compare the two attached PDFs. Make a table of every change in [terms, prices, dates, obligations], with the page number in each version, and mark which changes matter most.

Extract

Extract key facts and dates

Use in: Chat

Extract every deadline, amount, name and obligation from the attached PDF into a table with columns: item, value, who it applies to, page number.

Study

Turn a paper into study notes

Use in: Chat

Turn the attached paper into study notes: the research question, method, main results, limitations and 10 flashcard questions with answers. Explain any jargon simply.

Slides

Turn a PDF into a presentation

Use in: Agent · Slides

Create a [10]-slide presentation from the attached PDF for [audience]. One idea per slide, the key numbers as simple charts, and the source page in each slide's notes.

Translate

Translate and keep the layout

Use in: Chat

Translate the attached PDF into [language]. Keep the headings and lists, keep numbers and names unchanged, and list any terms that are hard to translate.

Review

Review a contract or policy

Use in: Thinking

Review the attached [contract] from the point of view of [my role]. List unusual or one-sided clauses with page numbers, explain each in plain English, and suggest questions to raise. This is not legal advice.

Want more? See the full Kimi prompt library.

COMPARE

Compare several PDFs at once

Attach two or more PDFs to the same chat and Kimi can line them up side by side. It works well for contract versions, research papers, supplier quotes and annual reports from different years.

  • 1
    Name the files. Clear names such as contract-v1.pdf make the answer easier to follow.
  • 2
    Say what to compare. List the things that matter, such as prices, dates, terms or results.
  • 3
    Ask for a table. One row per item, one column per file, plus page numbers.
  • 4
    Ask what matters most. Have Kimi rank the changes, then check those pages yourself.
PDFcontract-v1.pdf18 pages PDFcontract-v2.pdf21 pages Compare TermVersion 1Version 2Pages Contract length12 months24 monthsp.2 → p.2 Monthly price$4,000$4,000p.5 → p.5 Notice period30 days90 daysp.9 → p.11 Liability cap$1M$250Kp.14 → p.17 Governing lawEnglandEnglandp.18 → p.21 ⚠ 3 changes found · check each page Example data for illustration

Kimi lines up two contract versions and highlights what changed.

CHECK

How to check Kimi's answers

Moonshot reports that Kimi K3.1 Pro makes far fewer factual errors than K3 on its own hallucination test, but no AI reads documents perfectly. These four habits catch most mistakes.

Ask for page numbers

Make every answer point to a page, and quote the exact sentence for important claims. Then open those pages.

Do the last-heading test

Ask “What is the last heading in this document?” If Kimi gets it wrong, it may not have read the whole file. Split it.

Re-check the numbers

Numbers from tables and charts are the most likely to be misread. Compare totals and key figures with the PDF.

Ask what's missing

Add “Tell me if the PDF doesn't answer this” so Kimi says so instead of filling the gap from general knowledge.

SCANS

Scanned PDFs, charts and locked files

PDFs with real text work best. Kimi doesn't publish how it handles scans, so use these steps when a file is a photo of pages, full of charts or locked.

Check if it's a scan

Try to select text in the PDF with your mouse. If you can't, it's an image of text, and results will be weaker.

Run OCR first

Use the text recognition (OCR) feature in your PDF or scanner app, save a new PDF, and upload that instead.

Send charts as images

Kimi K3.1 Pro understands images. For a chart that comes out wrong, screenshot that page and ask about it directly.

Remove the password

Password-protected PDFs often fail to upload. Save an unlocked copy if you're allowed to.

PROJECTS

Use Projects for documents you reuse

Upload once

A Kimi Project keeps your reference PDFs, chats and instructions together, so you don't re-upload the same files in every chat.

Read on demand

Kimi picks the project files it needs for each question instead of loading every file every time.

Set instructions

Add standing instructions such as “Always give page numbers” or “Answer for a non-expert”.

Good for: a thesis with dozens of papers, a deal with many contracts, or a team policy library. For files on your own computer, Kimi Work can read whole folders.
API

Analyze PDFs with the Kimi API

To process many PDFs in an app or script, upload each file with the purpose file-extract, read the extracted text, and send it to kimi-k3.1-pro as a system message.

import os from pathlib import Path from openai import OpenAI client = OpenAI( api_key=os.environ["MOONSHOT_API_KEY"], base_url="https://api.moonshot.ai/v1", ) # 1. Upload the PDF and extract its text f = client.files.create(file=Path("report.pdf"), purpose="file-extract") text = client.files.content(file_id=f.id).text # 2. Ask a question about it reply = client.chat.completions.create( model="kimi-k3.1-pro", messages=[ {"role": "system", "content": text}, {"role": "user", "content": "List the 5 key findings with page numbers."}, ], ) print(reply.choices[0].message.content) # 3. Delete the file when you're done client.files.delete(file_id=f.id)
  • 1
    Don't send the file ID. The model can't read a file through its ID. Send the extracted text.
  • 2
    Don't paste base64. Sending the raw PDF as base64 uses far more tokens than extracting it.
  • 3
    Keep the document first. Put the PDF text before your questions so context caching cuts the cost of follow-ups.
  • 4
    Check for empty text. Extraction can fail on broken files, unsupported formats or files over 100 MB.

K3.1 Pro costs $2 per million input tokens and $10 per million output tokens, with cached input at $0.20. Setup and keys: Kimi API guide.

FIXES

Fix common PDF problems

Kimi says it can't read the file
  • Check the file opens normally on your computer.
  • If it's a scan, run OCR first and upload the new PDF.
  • Remove any password, then upload again.
  • Try saving it again as a standard PDF (“Print to PDF”).
The upload fails or is stuck
  • Make sure the file is under 100 MB. Compress it or split it if not.
  • Check your internet connection and refresh the page.
  • Try another browser, or the Kimi app.
Kimi only seems to read part of the PDF
  • Ask “What is the last heading in this document?” to test.
  • Split very long PDFs into parts, such as by chapter, and ask about each part.
  • On Max and Ultra plans, K3 extra-long chat handles up to 1M tokens.
The numbers or tables are wrong
  • Ask Kimi to quote the exact row and page it used.
  • Screenshot the table and send it as an image.
  • Ask for one table at a time instead of all at once.
Answers mix in outside information
  • Start with “Using only the attached PDF…”.
  • Turn off web search for that question.
  • Ask Kimi to say when the PDF doesn't cover something.

Other issues? See Kimi troubleshooting.

FAQ

Kimi PDF FAQ

Yes. You can attach PDFs in Kimi chat on the web or in the app and ask Kimi to summarise them, answer questions, pull out tables or compare files. Kimi also supports Word, Excel, PowerPoint, images, TXT and video.

Each file can be up to 100 MB, and you can upload up to 50 files at a time, according to the Kimi Help Center. Kimi doesn't publish a page limit; how much it can read at once depends on the model and your plan.

Yes, you can upload PDFs and ask questions on the free plan, with limited Kimi K3.1 Pro use. Agent tasks such as making slides or spreadsheets from a PDF use agent credits, and the full 1M-token chat context needs a Max or Ultra plan.

Kimi can understand images, but results on scans are less reliable than on PDFs with real text. Kimi doesn't publish details of how it handles scans, so for best results run OCR in your PDF app first, then upload the new file and check the numbers.

Yes. Attach the files in the same chat and ask for a comparison table with page numbers from each file. This works well for contract versions, research papers and supplier quotes.

Start with “Using only the attached PDF”, ask for page numbers and exact quotes, and tell Kimi to say when the document doesn't answer the question. Then check the key pages yourself.

Yes. Upload the file to the Files API with the purpose file-extract, read the extracted text, and send it to a model such as kimi-k3.1-pro as a system message. Each file can be up to 100 MB.

Treat any online AI tool with care. Avoid uploading documents you aren't allowed to share, remove personal details you don't need, read Kimi's privacy policy and delete chats and files you no longer need.

Try it with your own PDF

Open Kimi, attach a file and paste one of the prompts above.