Contracts and policies
Ask what the text says about deadlines, liability or termination, and ask for the relevant clause quoted with its page.
CAPABILITY
Chat with documents: answers from PDF and DOCX
Attach a file to a message and ask about it the way you would ask someone who has already read it. The model searches by the meaning of your question rather than by exact word matches, and it sees each passage together with the page it came from. You can also ask it to summarise the whole document or quote one specific page.
- 11
- document formats
- up to 7
- files per message
- 10 MB
- total attachment size
How it works
-
Attach a file to your message
Use the paperclip in the chat input. Up to seven files per message, 10 MB in total.
-
The file is processed automatically
Nothing to click: the document is split into passages, and each one keeps its page number. Send the same file to the same chat again and it is not processed a second time.
-
Ask about the contents
The model finds the relevant places in the text and answers from them. With several files in the chat, it can search across all of them, or you can name the one you mean.
-
Ask for a quote or a summary
Search is not the only option: it can read the whole document, or the text of one specific page — "quote page 12".
What it does
- Input
- A document plus your question in plain text
- Formats
- PDF, DOCX, XLSX, XLSM, PPTX, TXT, Markdown, CSV, TSV, JSON, HTML
- Files per message
- Up to seven per message, 10 MB in total
- How search works
- By the meaning of the question, not by exact word matches
- Page references
- Each passage is passed to the model with the file name and a page number or page range
- Reading in full
- Up to 20,000 characters of text at a time, 120,000 in total; beyond that the text is cut off
- Reading a single page
- You can request the text of one specific page of a document
- Several documents
- Search across every file in the chat, or one of them if you name it
- Where it's available
- Chat in the web version and in the Telegram mini app
What people use it for
Reports and research
Find the methodology section or a single figure inside a long PDF without scrolling through it by hand.
Spreadsheets and exports
Upload an XLSX, CSV or JSON file and ask what it contains, which fields it has and which values appear in them.
Decks and handbooks
Ask a question about a PPTX or DOCX file and ask for the passage the answer was built on.
Models for this task
Full catalog
Language
GPT-5.6 Sol
Flagship of the GPT-5.6 series. 1M token context.
In $0.0075
/
Out $0.045
/ 1K
Language
GPT-5.6 Sol Pro
GPT-5.6 flagship with pro reasoning mode for complex tasks.
In $0.0075
/
Out $0.045
/ 1K
Language
GPT-5.6 Terra
Balanced GPT-5.6 model: code, reasoning, agentic tasks. Half the price of Sol.
In $0.0038
/
Out $0.0225
/ 1K
Language
GPT-5.6 Terra Pro
Terra with pro reasoning mode: higher quality on complex tasks.
In $0.0038
/
Out $0.0225
/ 1K
Language
GPT-5.6 Luna
Fast and cost-efficient model of the GPT-5.6 series.
In $0.0015
/
Out $0.009
/ 1K
Language
GPT-5.4 Nano
Large language model for chat, writing and analysis
Free
Language
Perplexity
Last Perplexity model
In $0.0028
/
Out $0.0028
/ 1K
Language
GPT-5.5
The latest version of GPT-5. More powerful, faster.
In $0.0026
/
Out $0.021
/ 1K
Language
Auto Router Beta
Smart auto-router: automatically picks the best model for your prompt (beta)
In $0.0038
/
Out $0.0225
/ 1K
Language
Claude Fable 5
Anthropic's smartest model — Mythos class
In $0.015
/
Out $0.075
/ 1K
Language
Claude Opus 5
Anthropic's flagship for demanding agentic work
In $0.0075
/
Out $0.0375
/ 1K
Language
Gemini 3.1 Pro
Large language model for chat, writing and analysis
In $0.003
/
Out $0.018
/ 1K
What else it does
The things people usually discover only once they start working.
- Document reading and search are available on paid models that support search tools
- An uploaded document stays in the chat it was sent to
- Each document processed shows up as a separate entry in your balance history
Frequently asked questions
Which files can I upload?
PDF, Word (DOCX), Excel (XLSX and XLSM), PowerPoint (PPTX), plus TXT, Markdown, CSV, TSV, JSON and HTML. Up to seven files per message, 10 MB in total.
Which PDFs can be read?
Ones with an extractable text layer — files where the text can be selected and copied. That text layer is what the model receives, together with the file name and page numbers.
How do I know which page an answer came from?
Each passage is passed to the model with the file name and a page number or page range. To get the page into the answer itself, ask for it directly: "cite the page".
Which models can read documents?
Paid models that support search tools. If another model is selected, the chat says so in a separate message.
How is document processing billed?
Processing a file draws on your balance and shows up as a separate entry in the history. If your balance is short, the file is rejected before processing starts, and the message tells you how many tokens are needed.
Where is an uploaded document available?
In the chat it was uploaded to — the file belongs to that conversation. Send the same file to that chat again and it is picked up already processed.
Related capabilities
Try it on your own task
A starting balance is already on your account, no card required. You pay only for what you generate.