ChatGPT Enterprise PDF visual retrieval: images, charts, diagrams, and plan limits

ChatGPT Enterprise can process the text and embedded visual content of PDFs uploaded in supported conversations. That is materially different from text-only PDF retrieval, which ignores embedded images and is still the default behavior outside the Enterprise Visual Retrieval path.

QUICK REFERENCE

........

PDF path	What ChatGPT processes
Enterprise PDF uploaded in a supported chat	Text plus embedded images, charts, and diagrams
GPT Knowledge or Project file	Text-only retrieval
Free, Pro, Team, or Edu PDF	Text-based retrieval rather than Enterprise Visual Retrieval

........

WHAT VISUAL RETRIEVAL ACTUALLY CHANGES

Visual Retrieval extracts and interprets visual elements that would otherwise be skipped when a PDF is handled as text only. A chart, diagram, or screenshot embedded on the page can therefore become part of the answer instead of disappearing during ingestion.

··········

WHERE IT IS AVAILABLE

OpenAI currently documents PDF Visual Retrieval as an Enterprise capability. Availability also depends on how the PDF is attached: a file uploaded during a supported interaction can use visual retrieval, while some knowledge and project-file paths remain text-only.

··········

WHY THE UPLOAD PATH MATTERS

A PDF can contain exactly the same bytes and still be processed differently depending on where it enters ChatGPT. The practical distinction is not simply file type; it is file type plus plan plus upload surface.

··········

HOW LARGE MULTIMODAL PDFS ARE HANDLED

For large multimedia PDFs, OpenAI describes a retrieval workflow in which text and images are embedded and indexed. Once the working content exceeds the active context threshold, relevant text-image pairs are retrieved rather than keeping every page fully active at the same time.

··········

DATA STUDIOS CAPACITY ESTIMATE

OpenAI's Enterprise guidance notes that roughly 350 mixed text-and-image pages can fill a 110k-token working context. That is an operating estimate, not a hard page limit: dense pages, image-heavy reports, and OCR quality can move the practical threshold materially.

··········

LIMITATIONS THAT STILL MATTER

Visual retrieval does not guarantee perfect extraction of values from complex image-based tables or low-quality scans. For exact financial, scientific, or operational numbers, structured source data remains preferable to relying on a rendered table inside a PDF.

··········

WHEN ENTERPRISE VISUAL RETRIEVAL IS THE RIGHT TOOL

The strongest use cases are reports in which meaning is split across prose and visuals: earnings decks, technical manuals, research papers, slide-like PDFs, process diagrams, and dashboards exported to PDF.

··········

DATA STUDIOS