Jev costs LKR 13.99 per million input tokens and Free per million output tokens on Roar AI, billed in Sri Lankan rupees on one invoice. A judge model: typed answers about your data, each with a probability, in well under a second.
Prices are shown in LKR to make model costs easier to estimate and manage for Sri Lankan teams, with the published US dollar rate beside each.
Example: 1,000 requests with 3,000 input and 1,000 output tokens each cost LKR 41.96. Every rate is published and stays the same from one request to the next. See every model.
Rates as of 20 November 2026. Rupee prices are our published US dollar rates converted at LKR 333 to the dollar, our rate for November 2026 — the same rate your invoice uses.
jevIf you are using a large model as a judge, it is worth testing. Jev answers typed questions, such as picking an option or scoring against a rubric, typically in 70 to 500 milliseconds, and only the input is billed. That suits grading other models’ output, guardrail checks, triage and routing.
It does not write text, so it cannot replace a chat model. TypeSafe lists maths and counting, comparing dates, multi-step logic and double negatives among its weak spots, and its probabilities are not guaranteed to be consistent, so check a sample of its verdicts before you rely on them.
| Model | Input | Output | Choose it for |
|---|---|---|---|
| Jev | LKR 13.99 | Free | Grading and guardrail checks. |
| GPT-6 Luna | LKR 33.30 | LKR 167 | Sorting and routing when you also want written answers. |
| DeepSeek-V4.1-Flash | LKR 64.94 | LKR 260 | A low-cost chat model for coding and tool use. |
Prices per 1M tokens.
Jev is not a chat model and does not write text. You give it some state (a conversation, a document or 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.
Roar AI provides access to Jev through the Roar AI gateway, with usage billed in LKR and no foreign card needed. Developers can use Jev alongside other supported models without maintaining a separate integration for every provider.
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": {…}}'Requests to Jev are processed on international infrastructure, outside Sri Lanka.
Roar AI’s locally hosted model is separate and is labelled as such: if a workload has to stay in the country, Qwen 3.8 27B LK runs on our own server in Colombo.
Input is LKR 13.99 per 1M tokens; output is Free per 1M tokens. For example, 1,000 requests with 3,000 input and 1,000 output tokens each cost LKR 41.96. Rupee prices are our published US dollar rates converted at LKR 333 to the dollar, our rate for November 2026 — the same rate your invoice uses.
Yes. Jev usage is billed in LKR on one monthly invoice with every other model you use, so there is no foreign card or separate TypeSafe AI account to set up.
No. TypeSafe AI does not publish the weights, so it is only available through an API like this one.
Jev can read up to 64K tokens in one request.
Create a free account, get an API key and call Jev with the OpenAI SDK, billed in Sri Lankan rupees.