SemIf ยท Qwen3.5-4B
private betaAvailable models
A focused catalog for low-latency structured decisions.
JEV model router
An if statement that understands language. JEV answers four typed questions about each chat message in one call. A plain rule then picks where the message goes: a local model, a cloud model, an image model, or a web search.
Every message uses the same JEV config below. Personal data always stays on the local model.
Free demo. No sign-in needed. Rate limits apply.
The JEV config: 4 typed questions
The routing rule
Choose an example or type a message, then route it.
This session
Assumed cloud price $ per 1M tokensWhat the router saved across the messages you routed on this page.
Stayed local0%0 of 0 messages
Cloud tokens avoided0estimated
Cloud cost$0.00vs $0.00 if all went to the cloud
Routing time-median JEV call
Estimate: prompt tokens are characters divided by 4, plus 500 reply tokens per message. "Stayed local" counts the local model and follow-up questions. Image and search requests go to tools, not to the cloud model. The price is an assumption that you can change.
| Message | Route | Intended | Provider | Latency |
|---|
Hello world: one yes/no question
{
"status": "ready"
}API first
Create a key in your account, then call the classifier endpoint.
curl https://swiftest.ai/v1/classifier \
-H "Authorization: Bearer $SWIFTEST_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "swiftest-jev-small",
"state": "User message: Prove that there are infinitely many primes.",
"questions": {
"route": {"type": "choice", "instructions": "Pick the best destination.",
"criteria": {"local_model": "Simple tasks", "cloud_reasoning": "Hard coding or proofs"}},
"complexity": {"type": "score", "instructions": "How hard is it?",
"criteria": ["Trivial", "Easy", "Moderate", "Hard", "Expert"]},
"contains_pii": {"type": "noul", "instructions": "Does it contain personal data?"}
}
}'
Swiftest Labs
Open LabsMany Minds, benchmark tools, public sessions, and community features remain available.