Clef playground: run Cloudflare’s Clef online
Run Clef-flash below with no account: pick an example, edit the text, press Run. Sign in to compare Jev, Clef and Clef-flash side by side on the same input and questions, answer by answer.
Clef-flash 무료 체험
예시를 수정해 실행해 보세요. 계정은 필요 없으며 하루 3회 무료입니다.
시나리오
티켓을 적합한 팀에 배정하고, 긴급도를 평가하며, 이탈 위험을 표시합니다.
자유롭게 수정하세요. 모델은 여기에 입력한 내용을 대상으로 시나리오의 질문에 답합니다.
의사결정 모델 나란히 비교
같은 입력, 같은 질문으로 두 모델을 동시에 실행합니다. 어디서 답이 일치하는지, 각각 비용이 얼마였는지 확인할 수 있습니다.
시나리오
What you can do here
- Try Clef-flash without an account. The panel at the top runs Cloudflare’s 9B decision model on an example you can edit. Each run answers two or three typed questions with a full probability distribution.
- Compare models side by side. Signed in, send the same input and questions to Jev and Clef-flash — or any two of Jev, Clef and Clef-flash — and see answer by answer where they agree. This is the fastest way to decide which model to build on, and it is something neither TypeSafe nor Cloudflare offers on its own.
- Design your own questions. The playground has the same model picker plus a question editor, saved threads and question sets you can reuse from the API and batch runs.
Reading the results
Each answer card shows the model’s whole distribution, not just the winner. A Choice answer lists every option with its probability, a Score answer places the input on your scale, and a Noul answer is the probability that the statement is true. In the side-by-side table, “Agree” means both models picked the same top option, the same side of 50% on a yes/no question, or the same nearest level on a scale.
When the two models disagree, look at the probabilities before deciding which is right: a 52/48 split and a 97/3 split are very different disagreements. The cases where one model is confident and the other is not are the ones worth labelling by hand.
Three things to know about Clef
- It now reads long inputs to the end. On 2026-10-04 Workers AI read only about the first 2,048 tokens of the state and silently ignored the rest. Our retest on 2026-10-07 found a sentence at the very end of 20,000 characters, and inputs over the 65,536-token context are now rejected rather than cut. The comparison with Jev has both tests.
- The benchmarks are Cloudflare’s own. Cloudflare reports Clef ahead of Jev on intent classification and tool retrieval and behind it on deciding whether to call a tool. Nobody has reproduced those numbers yet. Jev vs Clef has the details.
- It is open weight. Clef and Clef-flash are on Hugging Face under Apache 2.0, so you can move to self-hosting later without changing your questions. How to get Clef lists every way to run it.
What it costs
| No account | With an account | |
|---|---|---|
| Clef-flash | 3 free runs a day, examples only | 1 credit (3 for long input or more than five questions) |
| Clef | — | 2 credits (6 for long input or more than five questions) |
| Jev | 3 free runs a day, examples only | 1 credit (3 for long input or more than five questions) |
| Side by side | — | The two models’ credits added together |
New accounts get 50 free credits, valid for 30 days, and failed evaluations are refunded automatically. After that, credit packs start at $9.90 for 5,000; see pricing for a calculator per model.
JevStation is independent and not affiliated with Cloudflare or TypeSafe. Clef is Cloudflare’s model; we run it through Workers AI.
Frequently asked questions
- Is the Clef playground free?
- Yes, to start. Clef-flash runs on this page without an account, 3 times a day on the example question sets. A new account gets 50 free credits, no card: a Clef-flash evaluation costs 1 credit, a Clef evaluation 2, and a side-by-side run costs the sum of the two models.
- What is the difference between Clef and Clef-flash?
- Clef is Cloudflare’s 27B decision model and Clef-flash its 9B model. Both take the same questions and return the same answer shapes. Clef is the larger model; by Cloudflare’s own latency figures Clef-flash is about five times faster. On JevStation Clef-flash costs the same credits as Jev and Clef costs double.
- Can I compare Clef with Jev on my own data?
- Yes. Signed in, the side-by-side panel sends the same text and questions to any two of Jev, Clef and Clef-flash at once, shows where their top answers agree or differ, and lists each model’s latency, tokens and credits.
- Does Clef generate text?
- No. Clef is a decision model: it answers Choice, Score and Noul questions with probabilities. For writing, summarising or reasoning in prose you still need an LLM.
- Can I call Clef from code?
- Yes. With a JevStation API key, send "model": "clef-flash" or "model": "clef" to /api/v1/systemone. You can also call Workers AI directly with a Cloudflare account; How to get Clef covers both.
Related
- Jev vs ClefCloudflare’s open-weight Clef against Jev: price, context, latency, benchmarks and when to pick each.
- Decision modelsJev, Clef, Clef-flash, Perplexity Decider and OpenAI Luna Decisions on the same inputs: accuracy, calibration, cost and latency.
- How to get ClefEvery way to use Cloudflare’s Clef: free trial, JevStation API, Workers AI direct, self-hosted — with prices and code.
Make it part of your pipeline
Sign up for 50 free credits, save your own question sets and call the same evaluation from the API.