DeepSeek's "Expert Mode" Now Accepts File Uploads — But Only Reads Text

Key update

DeepSeek (深寻) has rolled out a file-upload capability for its recently launched "Expert Mode," it has been reported. The upgrade is available on both the mobile app and the web client and, according to the product prompts, accepts up to 50 files per session with a per-file limit of 100 MB. The company says the feature supports a wide range of documents and images but only recognizes textual content.

What it does — and what it doesn't

Uploads are handled by OCR rather than by a native visual understanding model. IT之家 tested the feature and found the same: images and scanned documents are converted to text via OCR before the assistant can work with their content. So can the system "see" images natively? Reportedly, no — it cannot interpret visual scenes or non-textual elements beyond what OCR extracts.

Limits, privacy and wider context

The practical limits (50 files, 100 MB each) make the feature useful for bulk document review and Q&A over text-heavy materials, but not for tasks that require genuine visual reasoning. As Chinese AI firms iterate rapidly on product features, questions about data handling, storage and cross-border flows persist — especially amid global scrutiny and export controls on advanced AI hardware. It has been reported that DeepSeek’s update focuses squarely on text extraction, keeping the feature within current technical and regulatory comfort zones.