Cloudflare AI Search reaches general availability with multimodal embeddings and OCR

Cloudflare has announced that AI Search is now generally available, combining Workers AI, Vectorize, R2, and Browser Run into a fully managed index and retrieval pipeline. The update introduces native support for multimodal formats, including image embeddings, optical character recognition (OCR) for PDFs, and support for larger files up to 10 MiB.

Multimodal Retrieval and OCR

AI Search now preserves both visual and textual signals by embedding image pixels directly while retaining captions. This implementation uses Matryoshka Representation Learning (MRL) to keep storage efficient and search fast. Native multimodal retrieval is available with the Qwen3-VL-Embedding model, allowing queries to match visual details like texture and composition that captions might miss. For text-only models, AI Search converts query images to text using ToMarkdown to maintain basic multimodal support.

The service also now accepts PDFs up to 10 MiB (up from 4 MiB) and includes OCR capabilities to extract text from scanned documents. OCR is available to all accounts and is billed as image processing ingestion tokens.

Pricing and Availability

Billing for AI Search will start on November 1, 2026. The pricing model covers ingestion, storage, and queries, with no instance hours or capacity units required. The free monthly allotment on all Workers plans includes:

• Ingestion: 5M tokens (shared pool for text and images)
• Storage: 10 GB
• Queries: 1,000 semantic queries and 1,000 full-text queries

Paid rates include $0.75 per 1M tokens for base ingestion, $2.00 per GB-month for storage, and $0.75 per 1k semantic queries. Future updates will include support for video and audio processing and improvements to the keyword search engine scaling.