JEV 1.13 API: a model that returns decisions instead of text
TypeSafe AI's JEV 1.13 (typesafe/jev-1.13) is live on FastMetal. You send it a situation and a set of questions, and it returns a typed value and a probability for each question. It does not generate text. TypeSafe AI calls this a "System 1 model": fast judgments in a fixed shape.
It uses the same API key and balance as every other model on FastMetal, and each call is billed at one flat price.
What it does
- Three question types.
choicepicks one option from a list,noulreturns the probability that the answer is yes, andscoregrades on a scale. One request can mix all three. - Probabilities with every answer. You get a probability for each option and a confidence value, so an input that could go either way shows up as low confidence.
- Low latency. TypeSafe AI quotes 70 to 500 milliseconds. We sent a customer message in Japanese with three questions through FastMetal and measured 0.26 to 0.57 seconds per call, gateway included.
- No prose. It cannot chat, summarize or explain its reasoning. It fits jobs where the shape of the answer is known in advance: classification, routing and guardrails.
Calling it
JEV has its own endpoint; it does not use chat completions.
curl https://api.fastmetal.ai/v1/decisions/jev-1-13 \
-X POST \
-H "Authorization: Bearer $FASTMETAL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"state": "Customer message: My order never arrived. I want a refund.",
"questions": {
"intent": {
"type": "choice",
"instructions": "What does the customer want?",
"criteria": {"refund": "Wants a refund", "track_order": "Wants to know where the order is", "other": "Something else"}
},
"is_urgent": {
"type": "noul",
"instructions": "Does this need an urgent response?",
"criteria": {"true": "A deadline is close", "false": "Not urgent"}
}
}
}'
The response uses the same keys as questions. intent holds the chosen option and the probability of each option. is_urgent holds the probability that the answer is yes.
Pricing and limits
| Item | Value |
|---|---|
| Price | One flat price per accepted request |
| Request body | Up to 8 KB |
| Questions | Up to 16 per request |
| Context | 32,000 tokens |
TypeSafe AI bills by input tokens and does not charge for output. FastMetal caps the request body and sells each call at a single flat price, so the cost does not change with the length of the body or the number of questions. Rejected requests are not billed. As with every model, calls are paid from your prepaid balance. The current price is on the pricing page.
More
The request format and the limits are in the Decisions API documentation.