Cloudflare Clef · on iajev

Cloudflare Clef: try image and text decisions.

Test Cloudflare Clef, a decision model that evaluates your context against a defined set of answers. Add an image or text, choose the possible outcomes, and inspect their probabilities. Use the same model through the iajev API or MCP.

API reference Connect through MCP

Try a Clef decision

Estimated hold: 7 credits

Clef · 1 credit per started 100 input tokens. The estimated hold varies with text and images. Final charges never exceed the hold; unused credits return.

64K-token model window · text up to 1 MiB. Text, images and question share the window; overlong state may be truncated.

Up to 4 PNG, JPEG or WebP images. Each: 4 MiB and 16 MP; total: 8 MiB. Images are not retained in history or saved presets.

If you leave this page to sign in, reattach your images when you return.

Your next decision starts here.

Add context above, or try an example to get started.

How to test Clef online

Select a sample image or add your own text and images. Selecting a sample does not run the model.
Open the options to define your question and possible answers. Review the estimated credit hold before running.
Sign in and run Clef when available. Compare the probabilities, then adjust the context or choices and run a new decision.

The model, and what you can use here

27B parameters: Cloudflare’s open-weight decision model, released under Apache 2.0.
Typed answers: The iajev API and MCP support Choice, Noul and Score. This playground starts with Choice.
Images + 64K context: Combine up to four images with long text in a 65,536-token model window. Overlong state may be truncated by the model service.

Three image decisions to try with Clef

Assess a parcel photo

Use the box image and choose between intact, damaged and unclear. Compare another angle or add a customer description before deciding whether a person should review it.

Classify a product

Use the headphone image to compare product categories. Replace it with your own item while keeping the same choices to evaluate whether the classification holds.

Classify a shelf image

Use the shop image to choose a food category. Include an unclear option when several categories are visible; a category decision is not an inventory count.

Understand the cost before you run

iajev charges 1 credit per started 100 input tokens, with a minimum of 1. The hold is estimated from your text and images. Actual usage is charged up to that hold; unused credits return and iajev covers any underestimate.

Clef and Jev serve the same kind of decision

Jev remains the default on iajev and uses 1 credit per started 1,000 input tokens. Choose Clef explicitly to evaluate an alternative on the same task. Quality and latency depend on the workload; no universal winner is claimed here.

A structured answer still needs evaluation

Probabilities describe model estimates, not guaranteed correctness. Test against labeled examples from your own workflow before acting automatically. Clef is not a chatbot and does not generate a written explanation.

View credit plans

For example, 850 actual input tokens cost 9 credits if the reserved amount is at least 9. Text, question, choices and images all contribute. A smaller WebP file reduces transfer size; smaller pixel dimensions can reduce visual tokens. Check the live estimate for your input.

View credit plans

Use Clef through the iajev API or MCP

The example below makes a text-only Choice request. Keep your iajev API key on the server. Use a new idempotency key for each new decision and reuse it only when retrying the same request.

Clef API request

curl https://iajev.com/api/v1/decisions/run \
  -H 'Authorization: Bearer YOUR_API_KEY' \
  -H 'Idempotency-Key: clef-first-decision-001' \
  -H 'Content-Type: application/json' \
  --data '{
  "model": "clef",
  "state": {
    "message": "The customer received a damaged parcel."
  },
  "decision": {
    "type": "choice",
    "question": "Next step?",
    "choices": [
      "Review",
      "No action"
    ]
  }
}'

For images, add an images array containing data URLs or {content_type, base64} objects. Remote image URLs are not accepted. The model field must be clef; omitting it keeps Jev as the default.

In an MCP client, call list_models to check availability, then run_decision with model: "clef", state, decision and an idempotency_key. Include images for a visual decision. Model discovery is free; inference uses the same credit rules as the API.

API reference Connect through MCP

Cloudflare Clef: common questions

Can I try Clef for free?: Preparing input and selecting sample images are free. Running Clef requires a signed-in account and enough credits for the displayed hold. New accounts receive 10 signup credits; the number of trials depends on each input.
What images can I send?: Up to four PNG, JPEG or WebP images: 4 MiB and 16 megapixels per image, 8 MiB combined. Text, images and the question share the 65,536-token window.
Does Clef write an explanation?: No. This playground returns a choice and probabilities for your defined answers. It is intended for structured decisions, not free-form chat or image generation.
How is Clef different from Jev on iajev?: Jev remains the default text decision model at 1 credit per started 1,000 input tokens. Clef adds image input and a 64K context window at 1 credit per started 100 input tokens. Both support Choice, Noul and Score through API/MCP.
Are images saved in my decision history?: Images are not retained in iajev history or saved presets. Reattach them when repeating a decision; this does not make a claim about the model provider’s retention policy.