BastionGPT reads documents up to about 1,000 pages and 40 MB per attachment, and it reads the entire document rather than a sample of it. That capacity doubled in 2026, up from about 500 pages earlier in the year, and it continues to grow as AI models improve.
If your work involves large medical records, evaluation packets, or records review, here is what the limits are, why they exist, and how to get reliable results from even the largest files in a HIPAA-compliant environment.
What are BastionGPT's document limits?
Pages: up to about 1,000 pages per document
File size: up to 40 MB per attachment
Document types: PDF, Word, Excel, images, and more. Scanned, faxed, photographed, and handwritten documents are all readable.
For the complete list of supported file formats, see What types of documents are supported?
Why can't the AI read my larger document?
The page limit reflects how much information today's AI models can hold in working memory while staying accurate. Every AI model on the market shows a "lost in the middle" effect on very long documents: it stays sharp on the beginning and the end of the document, but details in the middle get fuzzy. Push past that point and the AI can start inventing details, such as a medication that never appears anywhere in the chart.
In most industries that would be an annoyance. In healthcare it is a patient-safety problem. So we treat the cap as a quality safeguard: we test each model against real clinical documents and set the limit at the upper end of what is clinically safe, not the upper end of what is technically possible.
Why does ChatGPT seem to handle bigger files?
Tools that accept very large files usually are not reading the whole file. They rely on a technique called retrieval (RAG): the system picks roughly 20 half-page excerpts it predicts are most relevant to your question and answers from those alone. It is blind to everything else in the document.
Retrieval works well for pinpoint questions like "When was the last visit?" But for the work healthcare professionals bring to us, summarizing a complete record, reviewing a history for an evaluation, or finding the one new detail in hundreds of pages of repeated notes, the answer is only as good as the ten or so pages the system happened to select. BastionGPT gives the AI your entire document, so within the page limit, nothing is skipped. If you are comparing HIPAA-compliant alternatives to ChatGPT for large medical records, that difference matters more than the number on the file-size limit. See how AI word limits compare across the major AI tools for a detailed breakdown.
How do I summarize a medical record over 1,000 pages?
Two approaches work well today:
Split the document. Divide a 1,500-page packet into two parts, for example, and have BastionGPT review each one separately.
Summarize in passes. Work through the record about 200 pages at a time with a running summary. Ask for a concise summary of the first 200 pages, then upload the next 200 and ask the AI to add anything new. Long medical records tend to repeat the same notes visit after visit, so the running summary stays short while still catching each new finding.
If you would like help building a workflow for your specific documents, we offer free 30-minute one-on-one sessions with a prompt engineer. Email [email protected] to set one up, and see Is there training documentation available for BastionGPT? for more resources.
Which plan should I choose for large documents?
Every plan uses the same highest-quality AI models; what differs is how much data the AI can work with at one time. Professional is designed for smaller everyday documents. Professional Plus and above support the full document capacity described in this article.
If records review or long evaluations are part of your regular work, choose Professional Plus. The free trial covers both Professional and Professional Plus, so you can test your largest real-world documents before you decide.
Will the page limit keep increasing?
Yes. The limit stood at about 500 pages earlier in 2026 and is about 1,000 pages today. Whenever a new frontier model is released, we evaluate it against roughly 200 real-world clinical use cases, including how it handles very large documents, and new models typically go live within about three weeks. When a model proves it can read more pages without losing accuracy in the middle, we raise the limit. You can read more about how we test and roll out models in What AI models power BastionGPT?
If a specific document has you stuck today, start a free trial and test it, or email [email protected] with what you are working on and we will help you find the best approach.