{
  "version": "2",
  "id": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a",
  "title": "Cloudflare's AI Search exits preview with usage-based billing starting November 1, 2026",
  "summary": "Cloudflare says AI Search is now generally available, with hybrid search as the default index method for new instances, multimodal embedding support, and usage-based billing that begins November 1, 2026 after a reminder email the week before.",
  "body": "Cloudflare has moved AI Search from preview to general availability, according to a changelog entry on its developer documentation site. The same entry states that usage-based billing for the service begins on November 1, 2026, and that Cloudflare will send a reminder email in the week before billing starts. The changelog does not give a separate launch date beyond the general-availability announcement itself, so the exact day the service became generally available is not explicitly stated in the supplied text.\n\nThe billing model described in the entry is usage-based and covers several distinct meters. Cloudflare lists included monthly ingestion, storage, semantic query and full-text query usage, and points readers to a Limits & pricing page for the rates and the included amounts. The changelog does not reproduce those rates or the size of the included allowances, so the practical cost of running an instance at a given volume cannot be determined from this evidence alone.\n\nOne of the more consequential defaults concerns retrieval. New AI Search instances use hybrid search by default, which the entry describes as combining semantic vector retrieval with full-text matching. Cloudflare notes that a different index method can be selected when an instance is created, meaning the default is a starting point rather than a fixed behaviour. For developers who have already built retrieval pipelines around a single method, the default matters mainly for new instances rather than existing ones, though the entry does not describe any migration path or behaviour change for instances created earlier.\n\nThe entry also addresses how embedding and reranking calls are billed. Workers AI embedding and reranking calls made by AI Search are now included in AI Search pricing, and Cloudflare states that these calls no longer appear on a Workers AI bill or in AI Gateway logs. Other kinds of calls are treated differently: generation, query rewriting and external providers continue to use the customer's own account and gateway. That split means the billing surface for a search deployment is partly consolidated and partly unchanged, and teams that reconcile spend across Workers AI and AI Gateway will need to account for the shift.\n\nOn the subject of model support, the entry lists two multimodal embedding models that AI Search can use: @cf/qwen/qwen3-vl-embedding-2b, alongside google-ai-studio/gemini-embedding-2. The entry further notes that through the REST API and the public endpoint alike, images can be part of search and chat requests, while a Supported models page is cited for readers seeking the full embedding model list. Nothing in the changelog covers how accurate, how fast or how costly these models are, nor does it specify the regions or plans under which they may be used.\n\nIngestion planning is also affected by changes to file handling. For scanned PDFs, optical character recognition is available on every account, the entry says. Up to 10 MiB is allowed for plain-text or code files, and for PDFs with OCR enabled, while a 4 MiB limit remains for PDFs without OCR and other supported formats. For file limits, Cloudflare points to a Data source page, and for OCR pricing to Limits & pricing, which means OCR is broadly available but is not described as free.\n\nOne respect in which instance creation has been simplified: the type field is optional when an AI Search instance is created. A website source is inferred by Cloudflare from an HTTP or HTTPS URL, the company says, or an R2 source from an existing bucket name. For the common cases of pointing at a site or an existing bucket, this removes a required configuration step, though what happens when a value is ambiguous, or when neither pattern matches, is not stated in the entry.\n\nTaken together, the changes describe a service that is being positioned for production use rather than evaluation. General availability, a defined billing start date, default retrieval behaviour, multimodal embeddings, account-wide OCR and simplified instance creation are the kinds of items that typically accompany a move out of preview. The changelog frames them as a set of linked changes rather than a single feature, and each has its own documentation page for details the announcement does not carry.\n\nFor freelancers and small studios building search or chat over a client's own content, the most immediate operational point is the billing transition date. Because usage-based billing begins November 1, 2026, and a reminder email is promised the week before, projects that ingest large document sets or run high query volumes have a defined window to estimate costs against the Limits & pricing page. The entry does not provide those numbers, so any estimate has to come from Cloudflare's pricing documentation rather than from this announcement.\n\nA second practical consideration is the change in where embedding and reranking costs appear. If those calls no longer show up on a Workers AI bill or in AI Gateway logs, then existing dashboards, alerts or client-facing cost reports that key off those sources will undercount AI Search activity unless they are updated. The entry does not describe any replacement reporting surface for those calls, which leaves a gap that teams will need to resolve with Cloudflare's own tooling or documentation.\n\nThe default switch to hybrid search is worth noting for anyone who has tuned retrieval behaviour in a prototype. Because hybrid search blends semantic vector retrieval with full-text matching, results for keyword-heavy queries may differ from a purely semantic setup. The entry states that a different index method can be chosen at creation time, so the default is avoidable, but it does not compare the methods or describe when one is preferable. That comparison is left to the Hybrid search documentation the entry references.\n\nMultimodal support and OCR availability widen the kinds of content that can be indexed. Images can be included in search and chat requests through the REST API and public endpoint, and scanned PDFs can be processed with OCR on every account. The size limits are explicit: 10 MiB for plain-text or code files and PDFs with OCR enabled, and 4 MiB for PDFs without OCR and other supported formats. Those ceilings are concrete constraints that ingestion pipelines should enforce before upload rather than after a failure.\n\nWhat the evidence does not establish is equally important. The changelog gives no performance benchmarks, no accuracy figures for the multimodal models, no regional availability, no statement about service-level commitments, and no pricing rates. It also does not say whether existing instances are migrated to hybrid search, whether the optional type field changes behaviour for previously created instances, or how OCR pricing is metered. Those are open questions that the linked documentation may answer but that this announcement does not.\n\nThe reminder email is the only stated notification mechanism ahead of billing. Cloudflare says it will send that email the week before billing begins, which places it in late October 2026. For freelancers who manage several client accounts, that single notification channel is worth planning around, since the changelog does not mention in-console warnings, grace periods or a way to opt out of billing beyond not using the service.\n\nA reasonable reading of the announcement is that AI Search is now a billable production service with a documented set of defaults and limits, and that the transition period before November 1, 2026 is the moment to verify costs and reporting. That reading is editorial interpretation rather than a claim in the source: the changelog itself presents the changes factually and defers specifics to linked pages. Teams should treat the linked Limits & pricing, Hybrid search, Supported models and Data source pages as the authoritative detail behind this summary.\n\nThe unresolved items are mostly about fit rather than availability. Whether hybrid search suits a given corpus, whether the two multimodal embedding models meet a project's quality bar, and whether OCR costs are acceptable at a given document volume are all questions the changelog raises but does not answer. Until those details are read from the referenced documentation and, where possible, tested against real content, the announcement should be treated as a change in status and billing rather than a recommendation for any particular workload.",
  "category": "dev",
  "language": "en",
  "datePublished": "2026-10-05T07:17:52.710Z",
  "dateModified": "2026-10-05T07:17:52.710Z",
  "eventDate": null,
  "sourcePublicationDate": "2026-10-01T00:00:00.000Z",
  "source": {
    "name": "developers.cloudflare.com",
    "url": "https://developers.cloudflare.com/changelog/post/2026-10-01-ai-search-generally-available/",
    "kind": "official-publisher"
  },
  "practicalImpact": "Editorial interpretation: teams running AI Search over client content should treat the period before November 1, 2026 as a cost-modelling window, checking the Limits & pricing page for ingestion, storage, semantic query and full-text query rates, and updating any cost dashboards that previously read embedding and reranking spend from Workers AI bills or AI Gateway logs, since those calls are now folded into AI Search pricing.",
  "limitations": "The changelog does not state the exact general-availability date, the pricing rates or included usage amounts, regional availability, service-level commitments, or performance and accuracy characteristics of the supported models. It does not say whether existing instances are migrated to hybrid search, how the optional type field behaves for previously created instances, or how OCR is priced. No independent testing or third-party verification is reported in the supplied evidence, and the linked documentation pages were not provided.",
  "keyPoints": [
    "Cloudflare says AI Search is generally available, with usage-based billing beginning November 1, 2026 and a reminder email the week before.",
    "New AI Search instances default to hybrid search, which combines semantic vector retrieval with full-text matching; a different index method can be chosen at creation.",
    "Workers AI embedding and reranking calls made by AI Search are included in AI Search pricing and no longer appear on the Workers AI bill or in AI Gateway logs, while generation, query rewriting and external providers still use the customer's account and gateway.",
    "AI Search supports the @cf/qwen/qwen3-vl-embedding-2b and google-ai-studio/gemini-embedding-2 multimodal embedding models, and search and chat requests can include images via the REST API and public endpoint.",
    "OCR is available on every account for scanned PDFs; plain-text or code files and PDFs with OCR enabled can be up to 10 MiB, while PDFs without OCR and other supported formats remain limited to 4 MiB."
  ],
  "review": {
    "status": "source-reviewed",
    "checkedAt": "2026-10-05T07:17:52.710Z",
    "method": "Automated comparison against retrieved source text; not independent fact-checking.",
    "correctionNote": null
  },
  "sources": [
    {
      "id": 1,
      "url": "https://developers.cloudflare.com/changelog/post/2026-10-01-ai-search-generally-available/",
      "publisher": "developers.cloudflare.com",
      "title": "Changelog",
      "publishedAt": 1790812800000,
      "fetchedAt": 1791184654450,
      "hash": "5cc58b487cb48b2ac4eaf42b884c99ec87e1e48a062a33673e42e435191388e4",
      "kind": "official-publisher"
    }
  ],
  "claims": [
    {
      "claim": "Cloudflare says AI Search is generally available and that usage-based billing begins November 1, 2026, with a reminder email the week before.",
      "source": 1,
      "id": "claim-1",
      "url": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a#claim-1"
    },
    {
      "claim": "New AI Search instances use hybrid search by default, combining semantic vector retrieval with full-text matching, and a different index method can be selected at creation.",
      "source": 1,
      "id": "claim-2",
      "url": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a#claim-2"
    },
    {
      "claim": "Workers AI embedding and reranking calls made by AI Search are included in AI Search pricing and no longer appear on the Workers AI bill or in AI Gateway logs, while generation, query rewriting and external providers continue to use the account and gateway.",
      "source": 1,
      "id": "claim-3",
      "url": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a#claim-3"
    },
    {
      "claim": "AI Search supports the @cf/qwen/qwen3-vl-embedding-2b and google-ai-studio/gemini-embedding-2 multimodal embedding models, and search and chat requests can include images through the REST API and public endpoint.",
      "source": 1,
      "id": "claim-4",
      "url": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a#claim-4"
    },
    {
      "claim": "OCR is available on every account for scanned PDFs, with a 10 MiB limit for plain-text or code files and PDFs with OCR enabled, and a 4 MiB limit for PDFs without OCR and other supported formats.",
      "source": 1,
      "id": "claim-5",
      "url": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a#claim-5"
    },
    {
      "claim": "The type field is optional when creating an AI Search instance, and AI Search infers a website source from an HTTP or HTTPS URL or an R2 source from an existing bucket name.",
      "source": 1,
      "id": "claim-6",
      "url": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a#claim-6"
    }
  ],
  "formats": {
    "html": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a",
    "markdown": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a.md",
    "json": "https://freelancenews.online/news/cloudflare-s-ai-search-exits-preview-with-usage-based-billing-d788314a.json"
  }
}