Cloudflare · Model guide
Clef
Cloudflare's decision model on Workers AI. It answers a set of typed questions about a shared input: yes/no predicates, single-choice and graded scores, each with a probability. Called through /v1/systemone or /v1/decisions, not the chat endpoints. Accepts text and images. In our tests on long inputs it could miss evidence placed near the beginning while still finding evidence at the end; the whole input is counted as input tokens either way.
Returns probabilities, not text. Call it through /v1/decisions or /v1/systemone.
What this model is good at
What the vendor positions it for, and which of our models to reach for instead.
Classification and routing at volume · Each call answers a set of yes/no, single-choice and graded questions with calibrated probabilities instead of generated text.
- cloudflare/clef-flashsibling —Clef Flash is the smaller sibling for short inputs.
- typesafe/jevsibling —Use Jev when evidence may sit anywhere in a long text.
Pricing and billing
Billed on input tokens only. There is no output or cache price.
| Input size | Input / 1M tokens |
|---|---|
| Any size | $ 0.24 |
The whole input counts, including question text and option descriptions. Output tokens, where the upstream reports them, are not charged.
What it answers
A decision model takes one shared input and a set of typed questions, and returns a probability for each.
- Context window
- 65.5K tokens
- Max output
- —
- Input / output modalities
- Text + Image → Decision
- Question types
- Yes / no · Single choice · Graded score
- Output
- Probabilities per question — no generated text
- Streaming
- Not applicable — one response per call
- Image input
- Supported
- Endpoints
- /v1/decisions · /v1/systemone
Frequently asked questions
What is Clef?
How much does Clef cost?
What kinds of questions can Clef answer?
Which endpoint do I call?
Can I try it without writing code?
Try it on your own input
Fill in an input and a few questions in the Playground and see the probabilities, then wire it in.