Use cases
@cf/cloudflare/clef and clef-flash; weights on Hugging Face (Apache 2.0).Clef is Cloudflare’s decision model line in the same family as TypeSafe Jev: state + typed questions, probabilities out, no free-form text.
The practical delta is multimodal state and Workers AI placement. If your judgment needs pixels, start here.
Last verified: October 2026. Background: Clef review, what is a System One decision model.
Urgent Noul + team Choice on ticket text. Cloudflare’s own launch examples start here for a reason: it is the hello-world of System One.
Fetch or render a page (for example with Browser Run), pass text or screenshot state, Choice over category. Cloudflare has published sub-3s workflows versus heavier generative models on the same path. Re-benchmark yourself; numbers move.
Score user submissions on your rubric. Probabilities drive auto-remove vs review. Keep policy text in the question criteria, not in a second generative rewrite.
Noul: “should I take this action?” Choice: allow / confirm / block. Tens-of-milliseconds class decisions matter when the agent sits on Workers or near the edge.
Up to four embedded PNG/JPEG/WebP images with the request (Workers AI limits apply). Moderate image posts, check screenshot claims, flag UI states. Text-only Jev/Laya cannot do this without a separate vision step.
Choice over component area or owning team from a bug screenshot plus title. Cuts “please describe the screen” loops in QA bots.
Pass page images plus a question like “is this a government ID?” or “does the stamp look present?” Still not a full KYC system; it is a fast filter.
Choice of downstream Worker, queue, or model tier from request metadata and body. Useful when the router itself should live on Cloudflare’s network.
Any high-volume action where p(yes) above a threshold runs automatically and the gray zone pages a human. Pattern is identical to Jev; ops differ (Workers AI keys, binding env.AI.run).
Run Clef / Clef-flash weights under Apache when data residency or custom fine-tunes matter. Cloudflare has discussed fine-tuning paths; treat launch posts as intent until your contract says otherwise.
| Model | Rough role | List $/M input (Oct 2026) |
|---|---|---|
| Clef-flash (9B class) | Default high-QPS router | ~$0.09 |
| Clef (27B class) | Harder or multimodal-critical calls | ~$0.24 |
Promote to full Clef when flash confidence is chronically middling on your gold set, not because the brand name is bigger.
| Guide | Why |
|---|---|
| Top 10 Jev use cases | Managed TypeSafe patterns |
| Top Laya use cases | Self-host open weights |
| Jev vs Laya vs Clef | Pick a vendor |
| Decision model vs LLM for routing | Architecture |
See frontmatter FAQ for schema-ready Q&A (also rendered on the live guide page).
3 curated tools below.

Cloudflare Clef and Clef-flash are open-source (Apache 2.0) System One decision models on Workers AI: multimodal state plus typed noul/choice/score questions return probabilities, with a Jev-compatible API shape and optional fine-tuning path.

Jev is TypeSafe's flagship System One decision model: send application state plus typed Choice, Score, and Noul questions, and get calibrated probabilities your code can branch on, without generating free-form text.
Laya Studio hosts the open-source Laya System One decision model behind a Jev-compatible API: send state and typed questions, get answers with probabilities in one forward pass, plus optional MCP tools and a guard endpoint.
| Tool | Best for | Pricing | Billing note |
|---|---|---|---|
| Cloudflare Clef | AI Agent | Freemium | Free Trial |
| Jev | Productivity | Paid | Paid Service |
| Laya | AI Agent | Freemium | Free Trial |
What is Cloudflare Clef used for?
Clef is Cloudflare’s System One decision model family for typed Choice, Score, and Noul answers. Common uses: support triage, threat and domain classification, trust and safety scoring, agent guardrails, and visual classification with up to four images.
How is Clef different from Jev for use cases?
Wire format is largely System One compatible, so many Jev patterns port by changing endpoint and model. Clef adds multimodal input (images, and video in the open weights story) and runs on Workers AI with open Apache weights.
When should I use Clef-flash vs Clef?
Use Clef-flash when latency and $0.09/M list pricing matter more than peak accuracy. Use full Clef (~$0.24/M list, Oct 2026) when the decision is higher stakes or multimodal quality matters. Benchmark on your labels.
Can Clef classify screenshots?
Yes. Workers AI accepts optional embedded images with the state. That unlocks UI bug routing, brand-safety screenshots, and document page checks that text-only decision models cannot see.
Is Clef good for coding agents?
Yes for fast guardrails and tool gates on the edge. For TypeSafe-centric MCP and SDK examples, also read our Jev coding-agents guide. Pick based on where the agent already runs.