Cloudflare AI Search: GA pricing, RAG and Web Search API
Cloudflare AI Search is generally available. Compare its usage-based billing with Web Search API and design a retrieval workflow with source checks.
3 min read

In this article
Cloudflare AI Search is generally available, and billing begins on 1 December 2026. Cloudflare moved the start from 1 November in a 2 November changelog post, which says pricing and included usage are unchanged. Its pricing covers ingestion, stored index data and queries. A free monthly allowance remains available across Workers plans.
AI Search retrieves from indexed content. Web Search API through AI Gateway instead connects applications to web-search providers. They solve different retrieval problems and should not be treated as interchangeable names for one product.
Sources checked on 10 November 2026. Availability and prices can change; confirm the linked primary sources before acting.
Which search system do you need?
Use an indexed corpus when you need answers grounded in a defined collection, such as product manuals or approved support documents. Use web search when the question depends on current public information beyond that collection. Some applications need both, with separate access and freshness rules.
A retrieval result is evidence for the model, not an instruction. Keep source text isolated from application policy. If a retrieved page asks the assistant to ignore its rules or reveal data, the application must not treat that as authorized behaviour.
| Need | AI Search | Web Search API |
|---|---|---|
| Approved internal corpus | Index selected documents | Not a substitute for private authorization |
| Fresh public sources | Depends on indexed updates | Search external web providers |
| Cost planning | Ingestion, storage, queries | Provider and gateway billing terms |
Understand the announced bill
The GA announcement lists base ingestion at $0.75 per million tokens, storage at $2 per GB-month, semantic queries at $0.75 per thousand and full-text queries at $0.10 per thousand. Image processing adds $0.50 per million tokens to ingestion. Every Workers plan keeps a monthly free allowance of 5 million ingestion tokens, 10 GB-month of storage, 1,000 semantic and 1,000 full-text queries. Plan limits also differ: Workers Free allows 100 instances, 100,000 files per instance and 500 crawled pages a day, while Workers Paid allows 1 million files per instance (500,000 for hybrid search). Confirm current limits before estimating a deployment.
A useful estimate separates initial indexing from monthly changes. A mostly static manual collection behaves differently from a continuously refreshed news archive. Include the cost of answer generation and any third-party model or search-provider charges outside the retrieval service.
Design a small retrieval pilot
Start with a bounded collection of documents you have permission to index. Assign each document an identifier, source URL, version and access scope. Prepare questions with known answers and a few questions the collection cannot answer.
Measure retrieval before adding a fluent answer generator. Check whether the correct passage appears near the top and whether stale versions are excluded. A polished answer can hide a bad retrieval result, so retain the passages and their source identifiers for review.
Preserve authorization and deletion
A shared index must not make one customer’s private document available to another. Apply authorization before returning passages and validate the returned document scope again before producing an answer. Do not rely on a prompt to enforce tenant separation.
Test document deletion and replacement. A deleted policy should disappear from retrieval and any application cache according to the documented lifecycle. The default sync interval is now 24 hours (it was 6 hours) and can be set from 15 minutes to 24 hours, so choose an interval that matches how quickly removed content must leave the index. Include an operational procedure for reindexing, rollback and investigating an answer based on outdated material.
Who should use Cloudflare AI Search?
Teams that want managed ingestion and retrieval for a defined corpus have a reason to evaluate AI Search. Teams that only need fresh public links may be better served by a web-search integration. A small static FAQ may need neither.
Use the Cloudflare profile for wider platform context. This article presents an architecture and acceptance plan; it does not claim a live RAG deployment or measured retrieval accuracy.
Sources and further reading
- AI Search GA — Cloudflare
- Limits and pricing — Cloudflare
- Web Search API — Cloudflare
- Billing postponed and default sync interval — Cloudflare
Image credits
Featured layout: Cloudsoft Zone, CC BY-SA 4.0. No product logos or screenshots are used.
Frequently asked questions
When does AI Search billing begin?
On 1 December 2026. Cloudflare postponed it from 1 November in its 2 November 2026 changelog; pricing and included usage are unchanged (checked 10 November 2026).
Is Web Search API the same product?
No. Web Search API accesses search providers through AI Gateway; AI Search operates on indexed content.
Does RAG guarantee correct answers?
No. Retrieval quality, source freshness, authorization and answer validation all matter.
Sources
- AI Search GA — Cloudflare
- Limits and pricing — Cloudflare
- Web Search API — Cloudflare
- Billing postponed and default sync interval — Cloudflare, 2026-11-02
Read next

Industry updateSaaS and Cloud3 min read
OpenAI models on Google Cloud: availability, billing and governance
OpenAI models on Google Cloud: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna in private preview in Model Garden. Billing, data boundaries and what is still coming.
New tools, once a week
One short email listing what joined the directory.
Comments
No comments yet. Be the first to join the conversation.
Leave a comment
Every comment is read by a moderator before it appears. Your email address is never published; it is used only to contact you about this comment.
More from Cloudsoft Zone

ComparisonSaaS and Cloud7 min read
S3, Cloudflare R2 or DigitalOcean Spaces: real storage costs
What object storage really costs a small SaaS: S3 with CloudFront, Cloudflare R2 and DigitalOcean Spaces priced on one workload, with egress traps explained.
By Nilesh Kushvaha

Architecture guideSaaS and Cloud9 min read
Stripe subscription billing architecture that survives webhooks
Design Stripe subscriptions for a SaaS that stay correct when webhooks repeat or arrive out of order: data model, verified endpoint, worker and access rules.
By Nilesh Kushvaha

TutorialDevOps5 min read
Production monitoring for a small SaaS: signals, alerts and recovery evidence you actually need
Build production monitoring around critical user journeys, application errors, queue progress and capacity signals, with actionable alerts and a verified delivery path.
By Nilesh Kushvaha