Clef by Cloudflare

# Fix list: Clef

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/cloudflare-clef, the October 2026 research run, assessed 1 October 2026. Grade B, 66.3 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 Clef: 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, 61 out of 100, up to 7.8 more on the total

Why it scored 61: Scored as a hosted model on Workers AI. cloudflarestatus.com is a Statuspage with component history (20). The incidents feed covered 18 September to 1 October, 31 incidents, none naming Workers AI or AI Gateway. The history page is paginated by script, so the rest of the 90 days went unread, and Clef itself is a day old (8 of 30). The model page lists Clef under Text Generation, whose default limit is 300 requests a minute, though Clef isn't named on the limits page (13 of 15). The errors page gives 429 two internal codes, 3036 for a spent free allocation and 3040 for 'Capacity temporarily exceeded, please try again'. We found no `Retry-After` documented. A call has no side effects, so a retry is safe (10 of 15). No SLA for Workers AI found (0). The model page carries no beta label (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, 40 out of 100, up to 7.5 more on the total

Why it scored 40: No x402, MPP or L402 on Workers AI (0). Per-token prices published without a login, $0.24 per million input tokens for Clef and $0.09 for Clef-flash, with neurons at $0.011 per 1,000 (20). Workers AI includes 10,000 neurons a day on the Free plan, and Cloudflare's Free plan needs no card per the 30 September check (20). The hosted route needs a person to sign up in a browser. The weights download from Hugging Face without an account, but other open-weight listings get no onboarding credit for that, so Clef doesn't either (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. Security & auth, 69 out of 100, up to 5.4 more on the total

Why it scored 69: Read for a model API. Cloudflare API tokens scoped to Workers AI permissions on an account and revocable, per the Cloudflare MCP listing's check of 1 October. Permission is per account, not per model (27 of 30). The Workers AI data page (21 April 2026) says customer content isn't used to train models without explicit consent and isn't kept unless you store it, and the launch post says Cloudflare doesn't read, store or train on Clef requests or responses outside the fine-tuning service (18 of 20). Clef is pitched at support triage and threat intelligence, both untrusted input, and we found no guidance on adversarial or injected content in the model page, the launch post or the model card. Self-hosting runs custom code from the model repository (3 of 15). The launch post's fine-tuning flow captures datasets through AI Gateway. We didn't check per-request logs for Workers AI or AI Gateway this run, and account audit logs record token changes per the Cloudflare MCP check (6 of 15). security.txt points to a HackerOne programme and a disclosure policy, per the same check. Certifications weren't checked this run (15 of 20).

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.

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

Why it scored 75: The model page publishes a JSON Schema for input and output, with `model` limited to `clef` or `clef-flash`, 1 to 64 questions and image limits, and the account API's `GET /ai/models/schema` returns a model's schemas. Cloudflare's public OpenAPI, updated 2 October, has only the generic `/ai/run/{model_name}` path (22 of 25). llms.txt for the developer docs (10). The page says what Clef does, 'turns a state and a schema of typed questions into decisions', and the launch post adds use cases, but neither says when not to use it (10 of 20). Typed question kinds behind a `type` enum, with question-count and image limits. State is free-form by design (13 of 15). Examples in JavaScript, Python and curl, and a Workers AI error table with internal and HTTP codes, though nothing specific to Clef (13 of 15). Model IDs carry no version, the Hugging Face repositories have commit history, and the Workers AI changelog we read has no Clef entry, its newest entry dated 16 June 2026 (7 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.

## 5. Agent ergonomics, 75 out of 100, up to 4.1 more on the total

Why it scored 75: Read as an API an agent calls for a decision, as with Jev, since Clef doesn't call tools or generate text. Answers are a probability per allowed option, so output stays small and the cost is the state, within 65,536 tokens. No batch or caching found for Clef (18 of 25). The caller fixes the output shape, with up to 64 questions and up to 4 images a call (18 of 20). Workers AI errors carry internal codes and messages, and a 429 tells a spent free allocation from a capacity limit. Nothing Clef-specific, such as which question failed validation, is documented (14 of 20). Calls are stateless and safe to retry, but there's no retry guidance beyond 'please try again' (15 of 20). Callable over REST and from a Worker through the `AI` binding. There's no Clef SDK, the body repeats the model name the URL already carries, and we found no documented way to point TypeSafe's SDKs, which post to `/v1/systemone`, at Clef (10 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.

## 6. Maintenance & community, 56 out of 100, up to 3.9 more on the total

Why it scored 56: Read for a model. Released on 1 October 2026 (30). We found no stated minimum notice for Workers AI model changes. The changelog's notice of 8 May aliased Kimi K2.5 to the pricier K2.6 on 30 May, 22 days later (4 of 12). Clef is too new to show churn, and Workers AI moved one model to a pricier successor in May (4 of 8). The Workers AI changelog we read has no Clef entry, and we didn't check Cloudflare's community or support channels (6 of 15). No Clef SDK. Clef is called through Workers AI's REST API and binding, and we didn't check Cloudflare's SDK versions this run (8 of 15). The Hugging Face repositories date from 30 September and carry custom code that local runs depend on. The vLLM pull request the launch post links (#57250) adds a Jev-like mode for DiffusionGemma, not Clef, and we found no Clef support in vLLM's main branch (4 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.

## 7. Transparency & trust, 89 out of 100, up to 1 more on the total

Made of editorial 77, provenance 100.

Why it scored 89: Weights and scoring code under Apache-2.0, following the Qwen base models. Training code and data aren't published (27 of 30). The Workers AI data page and the launch post agree, no training without consent and nothing kept unless you store it. Cloudflare's privacy policy and DPA exist, per the Cloudflare MCP check (26 of 30). The Workers AI changelog posts dated notices, such as the 8 May notice for Kimi K2.5, but we found no deprecation policy (10 of 20). Cloudflare's subprocessor page names 12 third parties and 16 group companies with locations, per the Cloudflare MCP check of 1 October. Where Clef inference runs isn't stated (14 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.

## 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: Workers AI incident history before 18 September, since the history page is paginated by script and the feed starts there
- unchecked: whether cloudflarestatus.com has a Workers AI component
- Whether Clef has its own rate limit or shares the Text Generation default of 300 requests a minute
- Whether output tokens are billed. The pricing table lists an input price only
- Whether Clef will get a vLLM path. The vLLM pull request the launch post links (#57250, merged into vLLM's main branch on 22 September 2026 and in no vLLM release yet) is Matt Mastracci's DiffusionGemma structured-read mode, not Clef support
- Whether TypeSafe's SDKs can be pointed at Clef. Cloudflare calls the API fully compatible, but the paths and model names differ
- Cloudflare's benchmark and latency claims against Jev are its own, and we didn't reproduce them
- The model card's summary gave a default max_length of 16,384 tokens for local runs, against 65,536 on Workers AI

## Weaknesses

- Released on 1 October 2026, with no service record and no entry in the Workers AI changelog we read
- No limitations, failure modes or prompt-injection guidance in the model page or the launch post
- No SLA found for Workers AI, and Clef shares the Text Generation default of 300 requests a minute
- Self-hosting loads custom code from the model repository
- No SDK of its own, and no documented way to point TypeSafe's SDKs at it despite the compatibility claim

## 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.

- Send `model` as `clef` or `clef-flash` in the body as well as the model in the URL. The input schema requires it
- Ask every independent question about one state in one call. Up to 64 are allowed
- Try `clef-flash` first for text-only triage. It costs $0.09 per million input tokens against $0.24
- Read the internal code on a 429. 3036 means the day's 10,000 free neurons are spent, 3040 means capacity, so retry later
- Treat an answer about user-supplied text as a judgement that hostile input can steer. Cloudflare publishes no guidance on it

## What the review panel asked for

- versioned hosted model IDs
- stated minimum notice period
- List the output price
- State whether errors are billed

## 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.