AI Search is now generally available

AI Search Goes General Availability: What It Means for Developers

Cloudflare has just announced that its AI‑powered search service is now generally available (GA). The platform, which blends Workers AI, Vectorize, R2, and Browser Run into a single, fully managed indexing and retrieval pipeline, has been in beta for over a year. The GA release brings a host of new features—most notably multimodal embeddings that support images, PDFs, and larger files—alongside a clear pricing model that starts billing on November 1, 2026, while keeping a generous free tier for all Workers plans.

A Unified Search Stack

At its core, AI Search is a one‑stop shop for building search experiences. Developers upload content to R2, which is automatically vectorized by Workers AI. The resulting embeddings are stored in Vectorize, and Browser Run provides a low‑latency retrieval layer that can be queried from any edge location. This architecture means that the heavy lifting—embedding generation, storage, and retrieval—is handled by Cloudflare, freeing developers to focus on the user interface and business logic.

The service is already in use internally: Cloudflare powers search on its own blog and developer documentation. The GA announcement signals that the platform is ready for production workloads of any size, from small personal sites to enterprise‑grade knowledge bases.

Multimodal Embeddings: Beyond Text

One of the most exciting additions is native support for multimodal content. Previously, AI Search could index images by generating a caption and embedding that text. The new approach embeds the raw pixel data directly, preserving visual details such as layout, color, and fine‑grained features. At the same time, the caption is retained for textual understanding, giving developers a richer retrieval signal.

The underlying technology is Matryoshka Representation Learning (MRL), which compresses high‑dimensional visual embeddings into smaller, more efficient vectors without sacrificing relevance. The Qwen3‑VL‑Embedding model is the first to support this native image retrieval, and the system automatically detects whether an instance’s model can handle images.

PDFs now benefit from optical character recognition (OCR), allowing the text inside scanned documents to be searchable. Larger files are also supported, expanding the use cases to include full‑length reports, white papers, and multimedia assets.

Pricing and Free Tier

Starting November 1, 2026, Cloudflare will begin billing for AI Search. The pricing model is straightforward: developers pay for the number of embeddings stored and the number of retrieval requests made. However, the free tier remains generous across all Workers plans, ensuring that hobbyists and small teams can experiment without incurring costs.

The move to a paid model reflects the growing demand for AI search and the resources required to maintain the infrastructure. It also provides a clear path for scaling, as teams can upgrade their plan when their search volume grows.

Practical Use Cases

• Internal Knowledge Bases: Teams can index internal documentation, code snippets, and policy documents, enabling quick retrieval of context‑specific answers.
• E‑commerce Search: Retailers can index product images and descriptions, allowing customers to search by visual similarity as well as keyword.
• Content Discovery: Media sites can surface related articles, videos, and PDFs based on both textual and visual cues.
• Compliance and Legal: Law firms can search through large PDF contracts and scanned documents, leveraging OCR to find clauses quickly.

Getting Started

1. Upload Content: Store your files in R2.
2. Vectorize: Use Workers AI to generate embeddings. For multimodal content, ensure the Qwen3‑VL‑Embedding model is selected.
3. Index: Store the vectors in Vectorize.
4. Query: Use the AI Search API or Browser Run to retrieve relevant results.

The documentation includes step‑by‑step guides and sample code in JavaScript, Python, and Go. Because the service is edge‑first, latency remains low even for large datasets.

Closing Thoughts

Cloudflare’s GA release of AI Search marks a significant milestone in the democratization of AI‑powered search. By abstracting away the complexity of embedding generation, storage, and retrieval, the platform empowers developers to build sophisticated search experiences quickly. The addition of multimodal embeddings opens new possibilities for visual search, while the clear pricing model ensures that the service can scale with demand.

Whether you’re building a knowledge base, an e‑commerce site, or a compliance tool, AI Search provides a robust foundation for delivering fast, relevant results to users.

---

TL;DR: Cloudflare’s AI Search is now generally available, offering multimodal embeddings for images and PDFs, a clear pricing model starting November 2026, and a fully managed pipeline that lets developers build fast, AI‑powered search experiences without infrastructure overhead.