Model facts on Vercel AI Gateway
Vercel AI Gateway began listing Jev on 16 September 2026. It is billed on input only; the gateway reports zero output tokens because Jev returns choices, scores, and probabilities instead of generated text.
- Model ID: typesafe-ai/jev
- Type: evaluation · provider: typesafe-ai
- Context window: 32,000 tokens
- Pricing: $0.042 / 1M input tokens
Calling Jev with the AI SDK
Import experimental_evaluate from ai and pass the gateway model ID. The function is marked experimental, so pin your ai package version and re-check the signature when you upgrade.
import { experimental_evaluate as evaluate } from 'ai';
const result = await evaluate({
model: 'typesafe-ai/jev',
state: 'The support agent issued a full refund to the customer.',
questions: {
refunded: { type: 'boolean', instructions: 'Was a refund issued?' },
},
});Vercel, TypeSafe direct, or jevmodel.org?
Vercel AI Gateway is the easy path if your app already deploys on Vercel and you want Jev on the same bill. TypeSafe direct gives the official console, SDKs, and version pins such as jev-1.13.0. OpenRouter serves it as typesafe/jev-1.13.
jevmodel.org is for testing decisions in a browser playground first and then calling one endpoint with prepaid packs from $9.90 — no Vercel team, OpenRouter credits, or TypeSafe account required.
FAQ
What is the Jev model ID on Vercel AI Gateway?
typesafe-ai/jev. Vercel lists it as an evaluation model from provider typesafe-ai with a 32,000-token context window and zero maximum output tokens.
Which AI SDK function calls Jev?
experimental_evaluate from the ai package. You pass model, state, and questions. It is not generateText or streamText, because Jev returns typed answers rather than text.
How much does Jev cost on Vercel?
Vercel lists $0.042 per million input tokens, the same as TypeSafe’s published rate, billed through your Vercel AI Gateway account.
Why is the question type “boolean” in the Vercel example?
Vercel’s AI SDK example uses type: "boolean" for a yes/no question. TypeSafe’s native API calls the same kind of question noul. Match the name to the API you are calling.
I don’t use Vercel. What is the quickest way to try Jev?
The jevmodel.org playground runs Jev in the browser with no account for the first 10,000 input tokens, and 100,000 more after sign-in. Then use https://jevmodel.org/v1/systemone with a jevmodel.org key.