Cloudflare has moved AI Search into general availability, adding native image retrieval, optical character recognition for scanned PDFs and support for larger files. The October 1 announcement positions the service as a managed indexing and retrieval pipeline for websites and internal collections.
Native image embeddings preserve visual information that may be lost when a picture is reduced to a caption. Cloudflare says supported multimodal models can match image queries against indexed images and text, while text-only models can still use a caption-based fallback.
Supported text files and PDFs can now be up to 10 MiB, compared with the previous 4 MiB limit. OCR allows scanned pages to enter the search pipeline even when the PDF has no extractable text.
For teams evaluating the service, better retrieval and generated answers are separate concerns: locating relevant content does not itself establish the accuracy of a final response. Cloudflare says billing will begin on November 1, 2026, with a free allotment across Workers plans. Full video and audio ingestion remain planned developments rather than features of this announcement.
1 comment