01
Triage customer messages
Compare how models route a support ticket, identify urgency and rate the impact of an incident.
Example question
Which team should investigate?
Billing · Technical · Sales
INDEPENDENT MODEL TRIALS & COMPARISONS
Try Cloudflare Clef.
Compare decisions.
Try Cloudflare Clef and Clef-flash on support tickets, invoices and risk decisions. Subscribe to compare them side by side with Jev, including structured answers, response time and cost.
No account needed for a free single-model run. Sharing required.
Available in the playground
- Clef
- Clef-flash
- Jev 1.13
Cloudflare Clef playground
See what you can testDecision results
See what each model decides.
Run your sample to compare answers, probability distributions, time and cost. Or open a recorded example below.
Clef
Usage, credits & sharing details
Sharing is required for free runs. By running this task, you agree that its context, questions and answers may be published after review. Remove personal and confidential information first.
Shared tasks, questions and results are saved for 30 days for administrator review. Published submissions remain until removed.
Choose Clef or Clef-flash, one model per run. Sharing is required. Up to 5 attempts a day per internet connection, resetting at midnight UTC. Availability is limited.
Free runs are submitted for community review. Subscribers choose whether to share and can recover results for 10 minutes.
BUILT FOR EVERYDAY DECISIONS
Test Clef on everyday decisions.
Start with a sample, then bring your own context. Define the possible answers and see where the models agree — or diverge.
02
Review invoice details
Test whether a model spots missing information and consistently classifies a purchase.
Example question
What kind of purchase is this?
Hardware · Software · Services · Travel
03
Assess operational risk
Give every model the same severity rubric and inspect which situations it would escalate.
Example question
Does this situation need escalation?
Yes / no · Risk category · Severity rating
From context to comparison in three steps.
Read the user guide- 01
Bring the context
Choose a public example or enter your own text or structured data.
- 02
Define the decision
Ask yes/no questions, provide a set of choices or define an ordered rating scale.
- 03
Inspect the differences
Run a free model, or select several with a subscription. Compare answers, probabilities, time and cost.
Loading community comparisons…
01 / YOUR CHOICES
Compare Clef, Clef-flash and alternatives.
Run Cloudflare Clef, Clef-flash and Jev here. Explore Perplexity and Strands Decider in our guides, with clear notes on which models we have tested.
| Model | Try here | Our examples | Learn more |
|---|---|---|---|
| Clef | Free trial and subscription | Tested | Clef guide ↗ |
| Clef-flash | Free trial and subscription | Tested | Clef-flash guide ↗ |
| Jev | Subscription | Tested | Jev guide ↗ |
| Perplexity Decider | Guide only | Not tested here | Perplexity Decider guide ↗ |
| Strands Decider | Guide only | Not tested here | Strands Decider guide ↗ |
THE EVIDENCE
Independent decision model benchmarks.
We tested Clef, Clef-flash and Jev on 60 sample tasks about customer messages, invoices and risk. Expected answers were fixed before testing, and each model ran each task three times.
View benchmarks and method- Synthetic sample tasks
- 60
- Models tested
- 3
- Model runs
- 540
These are our measurements on a fixed set of short English samples. Explore the method, mistakes and limitations before applying the results to your own work.
START FREE. COMPARE MORE.
A quick trial or a closer look.
Try one model with a shared task. Subscribe when you want private comparisons across models.
| What is included | Free trial $0 No account or card required. | Subscription $9.90/ month Cancel the next renewal anytime. |
|---|---|---|
| Available models | Available modelsClef and Clef-flash | Available modelsClef, Clef-flash, and Jev 1.13 |
| Models per run | Models per runOne model at a time | Models per runUp to 3, side by side |
| Sharing | SharingSharing required; published only after review | SharingSharing off by default; you choose when to share |
| Included usage | Included usageUp to 5 attempts per day per internet connection, subject to availability | Included usage30,000 credits per paid month |
| Try free | View plan |
Free attempts reset at midnight UTC. Subscription credits depend on the model and input size; unused credits expire at the end of the billing period. Model API prices are listed separately. Compare official API prices
BEFORE YOU START
Clef trials and comparisons: common questions.
Learn about the models, free trials, sharing and subscriptions.
Read the Cloudflare Clef guide Get in touchWhat is Cloudflare Clef?
Cloudflare Clef is a model for structured decisions. Give it context and questions with allowed answers, such as yes/no, a choice or an ordered rating. ClefBench lets you try Clef and Clef-flash in your browser and compare their decisions with Jev through a subscription.
Can I try it without an account?
Yes. Choose Clef and Clef-flash for a free single-model run. Free access allows up to 5 attempts per day per internet connection, resetting at midnight UTC, subject to availability. Sharing is required. You can also view the recorded example without making a new model call.
Will my free task be public?
Free runs require agreement to share the context, questions and results. They are saved for review and only published after a successful run and administrator approval. Unpublished submissions are retained for 30 days; published submissions remain until removed. Remove personal and confidential details before running.
What changes with a private comparison?
Subscribers can compare multiple models with sharing off by default. Private tasks are not submitted to the community. Completed results can be recovered for 10 minutes; this is not a permanent history. OpenRouter and the selected model provider still process your input.
How do monthly credits work?
A paid month includes 30,000 credits. The estimate is shown before you run. Credits are reserved first, then settled per successful model using actual input tokens, with a minimum of one credit per successful model. Failed models are not charged, unused reservations are returned to their original billing period, and credits expire at the end of that period.
Can I cancel the subscription?
Yes. Cancel the next renewal from your billing settings. Your current paid period remains available until its end. You can also manage payment details there; refund conditions are described on the pricing and terms pages.
Does higher confidence mean a better answer?
A confident answer can still be wrong. Models express confidence differently, so their numbers are not directly comparable. Use the answers, your own expected outcomes and our published evaluation method to judge whether a model fits your task.
Is ClefBench affiliated with Cloudflare?
No. ClefBench is an independent service for trying and comparing decision models. Your subscription covers ClefBench features and monthly credits; it is separate from a Cloudflare account or Workers plan.
YOUR CONTEXT. YOUR COMPARISON.
Try Cloudflare Clef on your own task.
Choose a public example or bring your own context. Start with one free model, then subscribe when you need side-by-side comparisons.
A recorded example is also available — no new model call needed.