PDFs in
Send a PDF with your question; get an answer about it.
- Pick a model that reads PDFs.
- Add the PDF as a part of your message.
- Choose how the PDF is read.
- Read the answer.
1. Pick a model that reads PDFs
Models that take files in lists the ones that read a PDF themselves; the sample below uses anthropic/claude-sonnet-4.5. A model that cannot is still served: the PDF is read first, by a parser, and the model gets its text (step 3).
2. Add the PDF as a part of your message
A message's content is a text, or a list of parts. To send a PDF, make it a list: one text part for your question, one file part for the PDF.
| Part | Holds |
|---|---|
{"type": "text", "text": "..."} | your question |
{"type": "file", "file": {"filename": "...", "file_data": "..."}} | the PDF: a public link, or the PDF itself as a data:application/pdf;base64,... URL |
{"type": "file", "file": {"file_id": "..."}} | a PDF you kept in Files: see Files and storage |
import json
import os
import urllib.request
API_KEY = os.environ["OPENDUNES_API_KEY"]
ADDRESS = "https://opendunes.com/v1/chat/completions"
MODEL = "anthropic/claude-sonnet-4.5"
PROMPT = "Summarise this document in two sentences."
FILENAME = "dummy.pdf"
PDF = "https://www.w3.org/WAI/ER/tests/xhtml/testfiles/resources/pdf/dummy.pdf"
body = {
"model": MODEL,
"messages": [
{
"role": "user",
"content": [
{"type": "text", "text": PROMPT},
{
"type": "file",
"file": {"filename": FILENAME, "file_data": PDF},
},
],
}
],
}
request = urllib.request.Request(
ADDRESS,
data=json.dumps(body).encode(),
headers={
"Authorization": "Bearer " + API_KEY,
"Content-Type": "application/json",
},
)
with urllib.request.urlopen(request) as response:
reply = json.load(response)
print(reply["choices"][0]["message"]["content"])
print(reply["usage"]["cost"])Send a link when the PDF is online, a file_id when you keep it in Files and use it often, and a data: URL when it is a file on your side. The whole request stays under 32 MB.
Don't send a file_id that is not yours, or whose file has expired. The request is refused with a 422.
3. Choose how the PDF is read
A PDF can be read by the model itself, or by a parser that turns it into text first. Name the parser in plugins when the default is not the one you want.
| You send | The PDF is read by | Charged as |
|---|---|---|
| nothing | the model, if it reads PDFs; otherwise the page-billed parser | the model's input tokens, or the pages |
"plugins": [{"id": "file-parser", "pdf": {"engine": "native"}}] | the model itself | input tokens, beside your words |
"plugins": [{"id": "file-parser", "pdf": {"engine": "cloudflare-ai"}}] | a text parser | no charge for the parsing; the text it finds is input tokens |
"plugins": [{"id": "file-parser", "pdf": {"engine": "mistral-ocr"}}] | an OCR parser, for scans | the pages it reads |
A scanned PDF holds pictures of text, not text. Name mistral-ocr for it.
4. Read the answer
The answer is the text in choices[0].message.content, as on any text request. usage.cost is the whole charge, the parsing included.
To send the answer back as a stream, see Streaming. To hold the answer to a shape, see Structured outputs. Every field of the request: Chat.
Errors
The body says why. What to do on every status: Errors and retries.
| Status | Means |
|---|---|
| 400 | the model takes no PDFs, and no parser was named for it |
| 402 | the balanceBalanceWhat you can spend, in DA.More on Balance and charges → cannot cover the request |
| 404 | no such model |
| 413 | the request is larger than the door takes |
| 422 | the file_id is not yours, or its file has expired |