Cloudflare AI Search billing separates retrieval costs from model charges

Cloudflare’s [October 1 release note](https://developers.cloudflare.com/ai‑search/platform/release‑note/) sets November 1, 2026 as the start of AI Search usage billing. Managed Workers AI embeddings and reranking are included in AI Search pricing, rather than appearing separately on the Workers AI bill or in AI Gateway logs. This is the scheduled billing start, following October’s general availability announcement.

[Current pricing](https://developers.cloudflare.com/ai‑search/platform/limits‑pricing/) counts ingestion on the final indexed chunks after parsing and chunking. Overlapping text counts in every chunk where it appears. The original file’s length is therefore not a one‑to‑one proxy for the ingestion meter. For scanned documents, OCR‑extracted text also counts as image‑processing tokens in addition to base ingestion.

The [release history](https://developers.cloudflare.com/ai‑search/platform/release‑note/) separately identifies generation, query rewriting and external model providers: they continue to use your account and gateway. The [pricing page](https://developers.cloudflare.com/ai‑search/platform/limits‑pricing/) confirms that their usage is billed and logged through the applicable service. An absent embedding entry in Gateway therefore does not establish that a retrieval workload was unused or free. Cloudflare’s San Francisco office entrance, photographed December 17, 2021. Archival company context. Photo: [HaeB](https://commons.wikimedia.org/wiki/User:HaeB) / [Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Cloudflare_office_entrance_area.jpg), [CC BY‑SA 4.0](https://creativecommons.org/licenses/by‑sa/4.0). Cropped; adapted image under the same licence.