Jev

Jev is not a chat model and does not write text. You give it some state — a conversation, a document, another model’s answer — and typed questions: pick one of these options, score this against a rubric, or how likely is this statement to be true. It answers all of them at once, each with a probability, typically in 70 to 500 milliseconds.

That makes it a fast, cheap replacement for using a large model as a judge: grading other models’ output, guardrail checks, triage and routing. Tools such as deepeval and Arize already use it that way, and only the input is billed. It comes from TypeSafe AI, founded by Diogo Almeida, formerly of OpenAI.

For scale: 1,000 requests of about 3,000 tokens in and 1,000 out cost $0.14 — a long prompt and a page of answer each. Every rate is published and stays the same from one request to the next. See every model.

Jev is billed in rupees, on the same monthly invoice as every other model your team uses — no foreign card, no separate TypeSafe AI account, and one key for all of it.

Prompts sent to Jev are processed outside Sri Lanka. If a workload has to stay in the country, Qwen3.8-27B (Colombo) runs on our own servers in Colombo.

TypeSafe says English is where Jev is most accurate. If you grade Sinhala or Tamil content with it, check a sample of its verdicts by hand first.

Send the state to judge and the questions to answer; typed answers come back. Read the docs.

curl https://api.roar-ai.com/v1/evaluate \
  -H "Authorization: Bearer roar_live_…" \
  -d '{"model": "jev", "state": "…", "questions": {…}}'