Cloudflare AI Search reaches general availability with multimodal support

What happened

Cloudflare announced that AI Search, its managed search index and retrieval pipeline built on Workers AI, Vectorize, R2, and Browser Run, has reached general availability. The product now includes native support for multimodal content, optical character recognition (OCR) for PDFs, and the ability to handle larger files up to 10 MiB. Billing will begin November 1, 2026, though a free tier remains available on all Workers plans.

Multimodal embedding and visual search

AI Search now embeds images directly rather than converting them to captions first. This preserves visual details—texture, color, composition, spatial relationships, and fine detail—that would be lost if compressed into text descriptions. The service uses Matryoshka Representation Learning (MRL) to keep embeddings efficient while maintaining search speed and manageable storage.

The implementation works in two modes. With multimodal-capable embedding models like Qwen3-VL-Embedding, query images are embedded directly by the model and land in the same vector space as indexed images and text. For text-only embedding models, AI Search converts query images to text using ToMarkdown and searches against the resulting caption, providing basic multimodal support to every model.

This approach enables searches difficult to express in words: finding images by visual description, locating visually similar results by example image, or combining both (for example, searching "a bird with similar markings" or "a bird perched on a leafy branch with plums"). The feature is useful for product discovery, screenshot matching, charts, diagrams, and scanned documents where visual characteristics matter.

PDF OCR and larger file uploads

AI Search now accepts text files (Markdown, HTML, CSV, JSON, and similar) and PDFs up to 10 MiB, an increase from the previous 4 MiB limit. For PDFs that are scanned images with no extractable text, OCR can be enabled. AI Search then reads text from each page before chunking and embedding. OCR is available to all accounts and is billed as image processing ingestion tokens under the new pricing model.

Pricing and billing model

Billing for AI Search begins November 1, 2026. Cloudflare charges for three components: ingestion (tokens based on embedding model choice), storage (per GB-month), and queries (per semantic or full-text query). The work of parsing, chunking, embedding with Workers AI models, keyword indexing, and reranking is included in the base cost. There are no instance hours, capacity units, or monthly minimums.

All Workers plans receive a free monthly allotment: 5 million ingestion tokens (covering any file type), 10 GB of stored data, 1,000 semantic queries, and 1,000 full-text queries. Pricing for paid usage is $0.75 per million base ingestion tokens, $0.50 per million image processing tokens, $2.00 per GB-month of stored data, $0.75 per 1,000 semantic queries, and $0.10 per 1,000 full-text queries. Embedding and reranking are free when using select Workers AI models; third-party models are billed separately.

What's next

Cloudflare plans to extend the ingestion pipeline to support full video and audio processing, allowing customers to search rich media assets. The keyword search engine is being refactored to scale better with large content stores, addressing current limits. The team is also developing simpler ways to enable AI Search and create indexes for websites already running on Cloudflare, making it easier for AI agents to discover and consume content.

Source

AI-assisted, verified against the source. This post was drafted with AI from the primary source linked above and checked automatically against it. For anything you depend on, read the original.