Cloudflare AI Gateway

# Fix list: Cloudflare AI Gateway

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/cloudflare-ai-gateway, the October 2026 research run, assessed 9 October 2026. Grade B, 66.8 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Cloudflare AI Gateway: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reliability, 66 out of 100, up to 6.8 more on the total

Why it scored 66: Hosted reading. www.cloudflarestatus.com is a Statuspage with AI Gateway, Workers AI and API as separate components (20). The incidents feed held 50 incidents from 21 September to 8 October 2026, none naming AI Gateway. The API component, which carries the REST routes, had an 18-minute minor incident on 7 October and delayed permission changes for four and a half hours on 30 September. The history page loads by script, so the rest of the 90 days went unread (10 of 30). Unified Billing traffic is limited to 200 requests per 60 seconds per gateway, the Cloudflare API to 1,200 requests per five minutes per token, and BYOK traffic has no gateway limit (15). Over-limit and over-budget requests return 429, the Cloudflare API returns `retry-after` and `Ratelimit` headers for its own limit, and retries with backoff and model fallbacks can be set per request or per gateway. No `Retry-After` was found for the gateway's own 429, and nothing separates a rate-limit 429 from a spend-limit 429 (11 of 15). No SLA naming AI Gateway was found. The Enterprise SLA lists service-specific SLAs for R2, Queues and Workers only (0). The core product carries no beta label, and Machine Payments is in beta (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 2. Payments & pricing, 52 out of 100, up to 6 more on the total

Why it scored 52: x402 is in beta on `POST /ai/run` since 30 September 2026 for four open models of 241 in the catalogue. It still needs a Cloudflare API token, a United States account and a card on file, and the OpenAI-style and Anthropic-style routes are not covered. We did not call it (12 of 40). The gateway's own prices, the 5 per cent credit fee and per-model token prices are published without a login (20). Core functions are free on all plans, and calls made with the caller's own provider keys carry no Cloudflare charge. We did not sign up to confirm that no card is asked for (20). A person creates the account and the first token in a browser, and the x402 route needs both (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 3. Agent ergonomics, 73 out of 100, up to 4.4 more on the total

Why it scored 73: Router, read on the model lines (tool use, structured output, caching, context, batch, SDKs, errors). Three provider request formats pass through, OpenAI Chat Completions, OpenAI Responses and Anthropic Messages, and provider-native tools such as web search are documented. Tool behaviour depends on the routed model (15 of 20). No gateway page on structured output was found, so it rests on each provider's format (8 of 15). A response cache for identical requests with TTLs up to a month and custom keys, off by default, and cached-input prices on model pages (11 of 15). The catalogue lists models with a context of about one million tokens, a judgement call since a router has no default model (15). Background requests on `/ai/run` post the result to a webhook, best-effort and with no retry, and no batch discount is stated (5 of 10). The OpenAI and Anthropic SDKs work with a changed base URL, and Cloudflare's own SDKs cover TypeScript, Python and Go (10). Status codes are documented page by page (401 with 10000 or 2009, 400, 404, 429, 503) with `cf-aig-step` and routing headers on responses, and no consolidated error reference (9 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 4. Security & auth, 75 out of 100, up to 4.4 more on the total

Why it scored 75: Router, read on the model lines. Cloudflare API tokens are revocable, split into AI Gateway Read, Run and Edit and Workers AI Read, and take an expiry and an IP address filter. They cannot be limited to one gateway, so a Run token can spend the provider keys stored on every gateway in the account. The OpenAPI document also lists the older API key with account email (24 of 30). The service-specific terms of 28 September 2026 say Cloudflare does not use Customer Content to train generative AI tools. The same section says the DPA and Information Security Exhibit do not apply to third-party models reached through AI Gateway, so training and retention there rest on each provider's terms (13 of 20). Prompts and responses are logged by default. Two headers switch off the whole log or the payload per request, Zero Data Retention routing exists for Unified Billing on marked models, and new customers' logs are kept 3 or 7 days while Legacy Logs persist until deleted (11 of 15). Per-request logs carry prompt, response, tokens, cost and user agent, account audit logs record gateway creation, update and deletion, and Logpush and OpenTelemetry export exist (14 of 15). security.txt names a HackerOne bounty programme and a disclosure policy, with no Expires field. The certifications page drew no list for our reader (13 of 20). Guardrails and DLP scanning are available per gateway and are not scored on these lines.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 5. Schema & documentation, 75 out of 100, up to 4.1 more on the total

Why it scored 75: Router, scored on the API lines. Cloudflare's public OpenAPI 3.0.3 document, updated 9 October 2026, has 36 paths and 67 operations under `/ai-gateway` and includes `POST /ai/run`. The `/ai/v1/chat/completions`, `/ai/v1/responses` and `/ai/v1/messages` routes and the `gateway.ai.cloudflare.com` endpoints are not in it (20 of 25). An AI Gateway `llms.txt` and a Markdown copy of every page (10). Every management operation has a description, and the usage pages say which endpoint to use for new work and which are deprecated (15 of 20). Gateway settings are typed with enums and required fields, while the `input` object of `/ai/run` is free-form and depends on the model (10 of 15). Curl, SDK and Worker examples on most pages. No table of gateway error codes was found. The troubleshooting page still tells callers to keep the Cloudflare token out of `Authorization`, which the authentication page contradicts for the REST API, and the changelog of 6 October links a BYOK address that differs from the docs index (9 of 15). A dated changelog with RSS. The API has no version of its own beyond `/client/v4`, and the move of new customers to Workers Logs on 24 September 2026 has no changelog entry (11 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 6. Transparency & trust, 73 out of 100, up to 2.4 more on the total

Made of editorial 51, provenance 95.

Why it scored 73: Closed service under the Self-Serve Subscription Agreement and service-specific terms with a section for Workers AI and AI Gateway. Each model page links the provider's terms, `ai-gateway-provider` is MIT and the OpenAPI repository is BSD-3-Clause (16 of 30). The service terms, the logging and Zero Data Retention pages, the DPA (version 6.4, 3 April 2026) and the privacy policy of 4 November 2025 were read. The privacy policy gives retention criteria, not periods. The terms take third-party models outside the DPA, while the sub-processor list names five model providers for AI Gateway, and the two are hard to square (17 of 30). No deprecation policy was found. Deprecated endpoints carry notices with no dates, and section 8 of the agreement lets Cloudflare modify or discontinue a service without notice (6 of 20). The sub-processor list, last updated 1 October 2025, names Google, Anthropic, OpenAI, xAI and Groq for AI Gateway with locations. The catalogue on 9 October 2026 also carries models from Alibaba, ByteDance, MiniMax and others that the list does not name (12 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Terms of service: read, states 7 of the 7 things a reader expects, and has 2 clauses that cost points (6 of 10)
- Privacy policy: read, states 7 of the 8 things a reader expects (9.3 of 10)

## 7. Maintenance & community, 75 out of 100, up to 2.2 more on the total

Why it scored 75: Router, read on the model lines. The newest changelog entry is 6 October 2026 (30). No minimum notice for API changes was found. The Universal Endpoint and the compat route are marked deprecated with a promise to keep working and no end date, and the error-code change of 6 October 2026 took effect on the day it was announced (5 of 12). No gateway endpoint was shut down in the last 90 days that we found. Thirteen dataset, evaluation and spending-limit operations are marked deprecated in the OpenAPI document with no changelog entry (5 of 8). A dated changelog with RSS and eleven entries between 5 August and 6 October 2026. The Discord community was not sampled (11 of 15). Replies on the SDK repositories were not examined (3 of 10). `cloudflare` for TypeScript v7.3.0 was tagged on 2 October 2026, Python is at v5.9.0 and Go at v7.12.0, and `ai-gateway-provider` 4.0.1 dates from 11 September 2026 (14 of 15). The OpenAPI repository was updated on 9 October 2026 (7 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## Deductions

Each comes off the total. A fixed and documented problem counts for less at the next check.

- 6 October 2026. Rejected-credential responses on `POST /ai/run` changed status (ElevenLabs 403, Vertex 500 and other providers' 402 became 401 with code 2009, and Unified Billing rejections became 503) on the day of the changelog entry, with no advance notice found. Three points (https://developers.cloudflare.com/ai-gateway/changelog/).

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: AI Gateway incident history before 21 September 2026, since the status history page loads by script and the feed starts there
- unchecked: Cloudflare's certifications. The compliance resources page showed our reader a heading and a link to the dashboard, with no list
- unchecked: whether Free plan signup asks for a card. The pricing page says core functions are free on all plans, and we did not create an account
- unchecked: whether the x402 route answers 402 as documented. We sent no request to api.cloudflare.com or gateway.ai.cloudflare.com
- unchecked: the provider pages, WebSockets API, Worker binding, custom providers, Cloudflare Access, evaluations and OpenTelemetry pages, which were not read
- unchecked: replies to issues on the SDK repositories and in the Discord community
- Section 2.2(e) of the Self-Serve Subscription Agreement bars introducing automated agents or scripts into the Services to generate automated requests or mine data. How Cloudflare reads that for agent callers of a gateway is not stated, and it matters before any probe is run
- Section 2.2.2 of the same agreement permits benchmark tests, asks that published results include what is needed to replicate them, and gives Cloudflare a reciprocal right
- The x402 price per request, network and asset are not on the Machine Payments page. They arrive in the `PAYMENT-REQUIRED` header
- Whether the Enterprise SLA's uptime promise covers AI Gateway. Its text speaks of serving Customer Content and names no AI product
- How the sub-processor list (five model providers for AI Gateway, last updated 1 October 2025) fits with the service terms, which take third-party models outside the DPA, and with a catalogue of 24 author paths
- The first release date of AI Gateway was not established. The status component dates from 31 May 2024 and the changelog page starts on 2 January 2025
- security.txt has no Expires field. It is recorded as valid because the provenance field has no value for an incomplete file
- Three Markdown addresses on www.cloudflare.com (the terms, privacy policy and service terms with `.md` appended, as its robots.txt describes) answered 404, so the HTML pages were read
- The lead was right about both hosts. Since 21 May 2026 Cloudflare recommends the REST API on `api.cloudflare.com` and marks the Universal Endpoint and the compat route on `gateway.ai.cloudflare.com` as deprecated for new integrations
- Workers AI is listed separately as `cloudflare-workers-ai`. It shares the `/ai/run` route, the API token and the terms with this listing, and its x402 entry rests on the same Machine Payments page

## Weaknesses

- API tokens are account-scoped. Any token with AI Gateway Run can send requests through every gateway in the account and spend its stored provider keys
- The service-specific terms say the DPA and Information Security Exhibit do not apply to third-party models reached through AI Gateway
- No SLA naming AI Gateway was found. The Enterprise SLA lists service-specific SLAs for R2, Queues and Workers only
- On 6 October 2026 rejected-credential responses on `/ai/run` changed to 401 with code 2009, or 503 under Unified Billing, announced the same day
- Unified Billing traffic is capped at 200 requests per 60 seconds per gateway, and a 5 per cent fee applies to credit purchases
- Section 2.2(e) of the Self-Serve Subscription Agreement bars automated agents that generate automated requests into the Services, which matters before any probe is run

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Use the REST API at `api.cloudflare.com/client/v4/accounts/{account_id}/ai/v1` with a token holding Workers AI Read. A token with only an AI Gateway permission returns 401 with code 10000
- Send `cf-aig-gateway-id` on every Workers AI (`@cf/`) call and on dynamic routes. Without it a dynamic route resolves against the default gateway and returns 404
- Set `cf-aig-collect-log: false` or `cf-aig-collect-log-payload: false` on sensitive requests. Logging of prompts and responses is on by default
- Treat a 429 as a rate limit or a spent budget and a 401 with code 2009 as a rejected provider key. Do not retry a 2009
- Set `byok_only` on the gateway or send `cf-aig-no-wholesale: true` to stop a request without a provider key falling through to paid Unified Billing credits

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.