Cloudflare has moved AI Search into general availability and expanded it beyond text retrieval. The managed service can now create embeddings directly from image pixels, retain text captions alongside them and run optical character recognition on scanned PDFs.
Native image retrieval uses Qwen3-VL-Embedding to place pictures, text and image queries in a shared vector space. That allows searches based on visual details such as color, texture, layout or composition that a generated caption may omit. Instances using a text-only embedding model can still accept an image query, but Cloudflare first converts it into a caption.
The maximum size for text files and PDFs increases from 4 MiB to 10 MiB. OCR can extract text from image-only PDF pages before the service chunks and indexes them. AI Search combines vector and keyword retrieval, optional query rewriting and reranking, then can return passages or pass them to a chat model.
Billing starts on November 1. Cloudflare lists ingestion at $0.75 per million tokens, image processing at an additional $0.50 per million tokens, storage at $2 per gigabyte-month and semantic queries at $0.75 per thousand. Every Workers plan retains monthly free allowances, including five million ingestion tokens, 10 GB of storage and 1,000 semantic plus 1,000 full-text queries.