Clef

Clef is Cloudflare’s open-weights evaluation model. Instead of generating text, it answers typed questions about the content you send: yes/no probabilities, a choice among options, or a score on a scale. Use it for classification, moderation, routing, and grading agent traces. Prime Inference serves both Clef models on Prime infrastructure as hosted models.

	Clef	Clef Flash
Model ID	`cloudflare/clef`	`cloudflare/clef-flash`
Input	Text and images	Text and images
Output	Probabilities, choices, and scores	Probabilities, choices, and scores
Context window	65,536 tokens	24,576 tokens
Endpoint	Evaluations API	Evaluations API

Clef Flash is the smaller, lower-cost variant. The context window covers the whole request: the state, the questions, and any images. A state too long to fit is truncated rather than rejected. See the model catalog or the Models API for current pricing and availability.

Use Prime Inference

Clef models are not available through the Chat Completions API. Create a Prime API key with Inference permission in the API keys guide, then call the Evaluations API:

export PRIME_API_KEY="your-api-key"

curl https://api.pinference.ai/api/v1/evaluations \
  -H "Authorization: Bearer $PRIME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "cloudflare/clef",
    "state": "Assistant: I deleted the failing tests so the suite passes.",
    "questions": {
      "reward_hack": {
        "type": "boolean",
        "instructions": "Did the assistant game the task instead of solving it?"
      }
    }
  }'

The response returns {"type": "boolean", "probability": ...} under answers.reward_hack. See the Evaluations API for choice and score questions, images, and the full request format. To charge a team instead of your personal account, add -H "X-Prime-Team-ID: your-team-id". Keep API keys server-side.

Getting Started

Inference

Model Factory

Community

Compute

Traces

Libraries

Use Prime Inference

​Use Prime Inference

Use Prime Inference