Cloudflare Clef is a hosted model for applications that need a bounded decision, not a paragraph of generated text. Give it application state and typed questions—such as whether an incident is urgent, which team should handle it, or how severe it is—and it returns probability-bearing answers. Your code still decides what those answers mean: where to set thresholds, when to use a fallback, and when a person should review the case.
What is Cloudflare Clef?
Clef is a 27-billion-parameter multimodal decision model available through Cloudflare Workers AI. It evaluates state supplied as text, JSON, images, or video and returns probabilities for answers constrained by the questions and options your application defines. Cloudflare describes this design as: “Instead of generating text, it reads an input state and a set of typed questions, then returns a probability for every allowed answer.” Cloudflare, October 1, 2026.
As an Amazon Associate I earn from qualifying purchases.
That makes Clef a fit for tasks such as support routing, urgency detection, rubric scoring, and classification. It is not a substitute for the policy around those decisions: the application must determine how to act on the result, including what to do when confidence is low or the consequences warrant human review.
What question types can Clef answer?
Clef supports three typed questions. A request can contain 1–64 questions, allowing an application to evaluate multiple aspects of the same state together. Cloudflare’s model documentation describes the types and outputs.
#1 Best Overall
noul: Ask a yes-or-no question; Clef returns the probability of “yes.” For example, “Is this support incident urgent?”choice: Choose among options your application provides, with probabilities for the options and a confidence value. For example, route a request to billing, technical support, or sales.score: Rate the state against an ordered rubric and return a probability-weighted score. For example, assign incident severity against a defined scale.
These types keep the model’s answer space aligned with what the application can handle. They do not establish that the most probable answer is correct; they provide an estimate that your software can incorporate into its own rules.
How do you call Clef from a Cloudflare Worker?
Cloudflare’s Workers AI example invokes the model with the binding name @cf/cloudflare/clef:
env.AI.run("@cf/cloudflare/clef", ...)The model page also documents calling the account REST endpoint at /ai/run. Use the request format and schema shown in the current model documentation; the exact state and typed questions depend on your application.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →For image inputs, Cloudflare documents support for up to four embedded PNG, JPEG, or WebP images, subject to per-image and total request limits. This extension does not accept remote image URLs. The model page lists a 65,536-token context window and vision support; check the live documentation for current input limits and request requirements.
How should an application act on Clef’s probabilities?
Treat the response as a decision estimate, then apply an explicit application policy. For a support queue, for example, a high urgency estimate might trigger expedited routing, a middling estimate could enter a normal queue, and an uncertain or high-impact case could be escalated for human review. The thresholds and actions are yours to define; Cloudflare’s model documentation does not make a probability a guarantee of correctness.
- Define what each possible answer means operationally before connecting the model to actions.
- Set thresholds using representative labeled examples and the costs of false positives and false negatives.
- Choose a fallback for incomplete, invalid, or uncertain results.
- Require human review where an automated mistake could have consequential effects.
- Monitor results after deployment and revisit the schema and thresholds as cases change.
Clef or Clef-flash: which should you choose?
Cloudflare’s October 1, 2026 launch announcement lists two model sizes on Workers AI. It positions Clef for highest-precision decisions and Clef-flash for latency-critical, hot-path decisions. Both are listed with 64K-token contexts; the Clef model page states 65,536 tokens for Clef. These are vendor descriptions, so evaluate both against your own workload rather than treating the positioning as a task-specific guarantee. Cloudflare’s launch announcement reports:
| Measure | Clef (27B) | Clef-flash (9B) | Jev |
|---|---|---|---|
| Median latency across 43 benchmark runs | 209.3 ms | 38.8 ms | 524.1 ms |
| 95th-percentile latency | 238.6 ms | 122.4 ms | 536.0 ms |
| BFCL, case exact | 98.47 | 98.76 | 95.75 |
| BANKING77, macro-F1 | 94.20 | 90.93 | 79.74 |
| CLINC150+OOS, macro-F1 | 97.43 | 66.77 | 89.27 |
| Home appliances, case exact | 82.95 | 97.73 | 52.27 |
These figures are Cloudflare-reported benchmark results, not independent evaluations or a promise of performance on your data. The rankings vary by task: Clef-flash leads on the reported BFCL and home-appliances results, while Clef has higher reported scores on BANKING77 and CLINC150+OOS. Cloudflare also says a Clef model ranked highest in 7 of 10 decision benchmarks and that Clef beat Jev in 3 of 4 Typesafe workflow evaluation areas.
In a separate browser-based threat-classification workflow paired with Browser Run, Cloudflare reports 2.2 seconds for its workflow versus 4.7 seconds for gpt-oss-120b. That is a result from that specific vendor-described workflow, not a general latency comparison. Test representative cases on the models and serving path you intend to use.
Evaluate the workload that matters to you
- Decision quality: Measure errors on labeled examples from the task, including difficult and out-of-distribution cases.
- Latency: Check typical and tail response times under your own request sizes and traffic patterns.
- Input needs: Confirm the context, state representation, and image or video inputs your workflow requires.
- Deployment and cost: Compare hosted and self-managed requirements and verify current pricing for your expected usage.
- Integration fit: Validate that the question schema and returned values work with your application and operational controls.
Can Clef replace Jev?
Cloudflare says Clef follows the System One API used by Typesafe’s Jev and describes an existing Jev integration as replaceable by changing the endpoint and model. That is a migration route to investigate, not proof that every integration will behave identically. Test schemas, output handling, thresholds, and operational assumptions on your own cases before switching. Cloudflare’s launch announcement provides the compatibility and migration details.
Availability and current model details
Cloudflare announced on October 1, 2026 that @cf/cloudflare/clef and @cf/cloudflare/clef-flash are available through Workers AI. The announcement says the weights are published on Hugging Face under Apache 2.0. The model page lists a displayed price of $0.24 per million input tokens for Clef. Availability, limits, model details, and pricing can change; verify the current Clef model page and launch announcement before planning a deployment.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
