A billing delay gives a search trial more time, but a slower default refresh changes when new documents can appear.
November 3 update: Cloudflare’s November 2 release note postpones AI Search billing from November 1 to December 1, 2026. The same note changes the default sync interval from six hours to 24 hours.
The new date changes the budget calendar
Cloudflare says AI Search usage will not be charged before December 1. Prices and included usage stay the same. Other providers and services used in the workflow can still have their own charges.
The pricing page includes 5 million ingestion tokens, 10 GB-month of storage, 1,000 semantic queries, and 1,000 full-text queries per month. Usage beyond those amounts is billed at separate rates.
| Usage beyond the allowance | Published rate |
|---|---|
| Base ingestion | $0.75 per million tokens |
| Image processing | An additional $0.50 per million tokens |
| Storage | $2 per GB-month |
| Semantic, vector and hybrid queries | $0.75 per 1,000 queries |
| Full-text queries | $0.10 per 1,000 queries |
For example, 100,000 hybrid queries use the 1,000-query allowance, leaving 99,000 billable queries. At $0.75 per 1,000, the query component is $74.25. That calculation excludes ingestion, storage, generation, and other applicable services.
A default interval is not a freshness guarantee
Cloudflare still supports selected sync intervals from 15 minutes to 24 hours. Teams can choose an interval in instance settings or through the sync_interval API field. The announcement describes a changed default; it does not say every custom setting is overwritten.
A sync interval and an answer’s correctness are separate properties. A document can be present in the index while an answer still misreads it. A correct answer can also rely on an older document when the index hasn’t refreshed.
Keep the CJK tokenizer check
The earlier October 16 keyword-matching update improves handling of Japanese, Chinese, and Korean queries through the trigram tokenizer. Changing that setting requires a full reindex.
Keep representative questions with known source documents. Compare retrieved passages before judging the generated answer, and verify the actual instance settings before changing the refresh schedule. A reindex also adds workload that should be separated from a normal month’s budget.
Review settings before the trial becomes a recurring bill
I would record the tokenizer, sync interval, and monthly query volume together. That makes it easier to explain whether a later problem concerns discovery, stale content, or answer generation. These are source-based recommendations; we have not run a retrieval benchmark.
Our Cloud Workload Security guide covers a different API boundary. Use the billing and sync announcement for the latest timing. How often does your document collection actually change?



