CLOUDFLARE has announced that AI Search is now generally available, expanding a platform that combines Workers AI, Vectorize, R2 and Browser Run into a managed index and retrieval pipeline. The GA release includes native multimodal embedding and retrieval, with support for image embeddings, OCR for PDFs, and the ability to handle larger files.
The system now preserves both caption-based understanding and native image signals, using Matryoshka Representation Learning to keep embeddings efficient while retaining rich visual detail. If an embedding model supports images, AI Search embeds query images directly into the same vector space as indexed images and text; if not, it can still search with an image by converting it to a caption via ToMarkdown.
This enables more nuanced searches across images, documents, charts and diagrams, applying either native image retrieval or caption-based querying as appropriate.
Pricing and practical use details accompany the GA launch. From 1 November 2026, billing for AI Search goes live, with a generous free tier retained across all Workers plans. Ingestion, storage and queries drive the cost, with the work of parsing, chunking, embedding, indexing and reranking included in the base charges. Ingestion is token-based and compatible across embedding models; an add-on applies if images are being processed.
Storage charges are per stored data, and query pricing differentiates semantic versus full-text searches, with a revised free monthly allotment of 1,000 semantic and 1,000 full-text queries. The article also signals plans to extend multimodal support to video and audio, and to streamline index creation for websites running on Cloudflare.