The News, Explained
Cloudflare made AI Search generally available on October 1, 2026. For website and internal-document search, it manages indexing and retrieves relevant material. Billing starts November 1. Source
Match processing to your files
Direct image retrieval needs an image-capable embedding model, currently Qwen3-VL-Embedding. Embeddings turn material into numbers for comparison. With a text-only model, a query image becomes a caption before searching. Source
Text files and PDFs can now reach 10 MiB, up from 4 MiB. Enable OCR to extract text from scanned PDFs. Every account can use OCR, billed as image-processing ingestion tokens under the new pricing. Source
Count each search allowance separately
All Workers plans receive 5 million monthly ingestion tokens shared across supported file types, plus 10 GB of storage. Semantic search, which compares meaning, gets 1,000 monthly queries; full-text search, which finds words, gets another 1,000. Operators can therefore check search type alongside total volume. Source
Charges cover ingestion, storage and queries. Embedding and reranking with select Workers AI models are included; third-party models are billed separately. File type, model and search mode together determine the applicable conditions. Source
OYOPICK’s Take
Start a document-search decision by matching the material to the retrieval method. Scanned PDFs use OCR to read text; product photos use image-capable models to compare visual features. Text extraction and visual similarity are different tasks, so check results on your own material before assuming comparable quality. Separate free query allowances could provide a useful starting point for understanding usage in a small collection.