RAG budget and freshness
Cloudflare AI Search: December billing and 24-hour sync
Cloudflare postponed AI Search billing to December 1 and changed default sync to 24 hours. Work through a cost scenario and check answer freshness.
In this article
The postponement prevents AI Search usage charges before December 1. It neither makes every associated service free nor guarantees timely document updates.
The two November 2 changes#
On November 2, 2026, Cloudflare announced that AI Search billing starts December 1 instead of November 1. The default sync interval also changes from 6 to 24 hours; it remains configurable from 15 minutes to 24 hours.
Documentation: Cloudflare · AI Search billing and sync, 2 November 2026
What a search budget includes#
The table lists published overage rates above monthly account allowances. It is not a complete bill: generation, query rewriting and external providers retain their applicable billing. Ingestion counts the final indexed chunks, including repeated text from overlap.
| Resource | Included per account per month | Overage |
|---|---|---|
| Base ingestion | 5 million tokens | USD 0.75 / million |
| Storage | 10 GB-month | USD 2.00 / GB-month |
| Semantic queries | 1,000 queries | USD 0.75 / 1,000 |
| Full-text queries | 1,000 queries | USD 0.10 / 1,000 |
Documentation: Cloudflare · AI Search limits and pricing
A worked scenario after billing starts#
Assume 20 million base-ingestion tokens, 20 GB-month stored and 10,000 semantic queries in an account with no other usage. Subtracting the relevant allowances gives USD 11.25 ingestion + USD 20 storage + USD 6.75 queries = USD 38. This excludes OCR, full-text, generation and other charges.
ingestion = max(0, 20_000_000 - 5_000_000) / 1_000_000 * 0.75
storage = max(0, 20 - 10) * 2.00
queries = max(0, 10_000 - 1_000) / 1_000 * 0.75
total = ingestion + storage + queries
assert total == 38.00
print(f"USD {total:.2f}")Documentation: Cloudflare · AI Search limits and pricing
How to check RAG freshness#
- Define acceptable delay for each document type before choosing the interval. A pricing page and a historical manual may need different policies.
- After changing a test document in staging, observe when a query returns the new version; retain edit, sync and indexing times.
- Check deletions, ingestion failures and citations as well as new documents. Successful sync does not prove an answer uses the correct version.
- Separate indexing, storage, retrieval and generation budgets. Recalculate from observed usage, chunk overlap and allowances shared across the account.
Before December, check associated resources that can incur charges too. The AI Search announcement does not automatically pause other services’ billing.