Does Grok 4.3 Have OCR?

That distinction matters if you are writing product copy, choosing a tool for expense processing, or building a workflow where misread numbers, dates, or merchant names could create real problems.

What the official xAI docs actually support

The strongest evidence here comes from xAI’s own documentation:

1. xAI says Grok can “search through and reason over” documents attached to chat messages. The documentation also says users can reference public files by URL or upload private files and reference them by ID, with the system automatically activating the attachment_search tool.^[2]
2. xAI’s Grok model page lists model capabilities including Text, Images, and Video.^[4]
3. xAI has a dedicated Image Understanding documentation page, indicating that Grok can work with image inputs in some form.^[13]

Those claims support a careful statement such as: Grok can use attached documents as context and has image-understanding capabilities. They do not support the stronger claim that Grok 4.3 has officially documented OCR or receipt-parsing support.^[2][4][13]

Why “image understanding” is not the same as OCR

Image understanding is a broad capability. It can mean a model is able to interpret visual content, describe what appears in an image, identify objects, or reason about what is shown.

OCR — optical character recognition — is narrower and more testable. It means extracting visible text from an image, ideally while preserving details such as:

• line order;
• columns and tables;
• merchant names;
• dates;
• totals and subtotals;
• decimal points;
• tax fields;
• item descriptions.

Receipts and scans are especially unforgiving. They can include small fonts, poor lighting, skewed angles, glare, folded paper, faded ink, multi-column layouts, handwritten notes, and confusing date or currency formats. A model may be able to understand an image generally without being officially documented as a reliable OCR engine.

What not to infer from the docs

In other words, the official documentation is enough to say Grok can work with documents and images. It is not enough to say Grok 4.3 has confirmed OCR support.^[2][4][13]

Third-party posts are not official OCR documentation

The source set also includes Threads, Hacker News, third-party articles, X posts, and YouTube videos discussing Grok 4.3 beta, document generation, PDF handling, or chat exports.^{[5][6][7][8][9][10][11][12]}

Those references may show what people are discussing or testing. They may also be useful leads for hands-on evaluation. But they are not xAI’s official OCR documentation, and they do not establish that xAI has formally supported photo OCR, scan text extraction, or receipt parsing in Grok 4.3.^{[5][6][7][8][9][10][11][12]}

For public product pages, sales material, compliance documentation, or internal procurement notes, it is safer to rely on what the official documentation actually says.

Better wording to use

A careful description would be:

According to xAI documentation, Grok can search and reason over attached documents, and xAI documents image-related capabilities including Image Understanding.^[2][4][13]

A claim to avoid would be:

Grok 4.3 officially supports OCR from receipts, scanned documents, and photos.

The second sentence goes beyond the supplied official evidence.

If you still want to test Grok for text extraction

You can run a practical evaluation, but treat it as your own benchmark — not as proof of an official OCR guarantee. A reasonable test set would include:

1. clear photos of printed documents;
2. low-light photos;
3. skewed scans;
4. long receipts;
5. small-font receipts;
6. tables and multi-column layouts;
7. handwritten notes;
8. documents with totals, dates, tax lines, and currency symbols.

Ask the model to output the text line by line, mark uncertain characters, and preserve structure where possible. Then compare the result with a human-verified transcript. Pay special attention to missing digits, decimal points, dates, merchant names, and field placement.

For reimbursement, accounting, audit, or compliance use, keep a human review step in place or use a tool whose OCR or document-extraction capability is explicitly documented.

Bottom line

Grok has officially documented document-search/reasoning and image-understanding capabilities.^[2][4][13] But the supplied official evidence does not confirm that Grok 4.3 can directly perform OCR on photos, scanned documents, or receipts.^[2][4][13]

The safest conclusion: say Grok has document and image capabilities; do not say Grok 4.3 has officially confirmed OCR support.