Back to AI decision models
Model comparison
Clef vs Clef-flash
Compare Clef and Clef-flash using the same provider-sourced ai decision models rubric. No mystery score and no invented benchmark ranking.
Quick take
Clef
Cloudflare's larger decision model for classifying information and choosing an action without generating a chat response.
Best for
- Support ticket classification and routing
- Image-assisted document checks
- Agent decisions with human escalation
Watch out for
This is not a chatbot and does not explain its decisions in prose. Test probability thresholds on your own examples. Hosted requests can truncate long state, and self-hosted limits depend on configuration.
Clef-flash
A smaller Cloudflare decision model for frequent routing and classification tasks, with a lower input-token price than Clef.
Best for
- Support ticket classification and routing
- High-volume structured decisions
- Agent decisions with human escalation
Watch out for
This is not a chatbot and does not explain its decisions in prose. Test probability thresholds on your own examples. Hosted requests can truncate long state, and self-hosted limits depend on configuration.
Compare the published facts
Clef vs Clef-flash
Values use each provider's own published units and limits. A blank means the provider did not publish a directly comparable value in the sources reviewed.
| AI decision models | Clef | Clef-flash |
|---|---|---|
| Input priceCurrent provider price for the text or structured state submitted to the model, including whether decision outputs are billed. | $0.24 / 1M input tokens | $0.038 / 1M input tokens |
| Input capacityMaximum documented state and question capacity for one request. | 65,536 tokens; up to 64 questions per request | 24,576 tokens on Workers AI; up to 64 questions per request |
| Decision typesThe bounded question forms the model can answer, such as choices, scores, or probabilities. | Choice, Score, and Noul probability questions | Choice, Score, and Noul probability questions |
| Output contractHow the response is constrained for direct use by application code. | Typed JSON decisions defined before inference | Typed JSON decisions defined before inference |
| UncertaintyWhether the endpoint exposes probabilities, confidence, or another calibrated signal. | Probability distributions; Choice and Score also include confidence | Probability distributions; Choice and Score also include confidence |
| Where to use itCurrent API, SDK, playground, and access-stage information. | Workers AI API and self-hosted weights | Workers AI API and self-hosted weights |
How to choose
Compare the job, not the hype.
Start with the job you need to complete, then validate cost, access, and policy details on your exact provider route.
Clef
Cloudflare says Workers AI customer content is not used for training or service improvement without explicit consent. Connected storage services can retain content. Self-hosted deployments follow your own storage and access controls.
Provider and API links
Clef-flash
Cloudflare says Workers AI customer content is not used for training or service improvement without explicit consent. Connected storage services can retain content. Self-hosted deployments follow your own storage and access controls.
Provider and API links
Frequently asked questions
Clef vs Clef-flash FAQ
What is the main difference between Clef and Clef-flash?
Clef: Cloudflare's larger decision model for classifying information and choosing an action without generating a chat response. Clef-flash: A smaller Cloudflare decision model for frequent routing and classification tasks, with a lower input-token price than Clef.
Should I choose Clef or Clef-flash?
Consider Clef when your priority is Support ticket classification and routing. Consider Clef-flash when your priority is Support ticket classification and routing. Test both with your own data and provider route before committing.
Is this Clef vs Clef-flash comparison based on Cody benchmarks?
No. This comparison aligns provider-published facts for the AI decision models category. It does not claim a universal winner or combine incompatible third-party benchmark scores.