Vercel AI Gateway: Jev Model Page

No. The card's own example is experimental_evaluate with a boolean question and model typesafe-ai/jev, and the model type is evaluation. Evaluation is not available through the OpenAI-compatible endpoint, so chat-completions code will not work against it.

Maximum output tokens is 0. It returns typed answers, scores and boolean probabilities rather than generated text, so the listed price is $0.042 per 1M input tokens against a 32,000-token context window. There is nothing to bill on the response side.

The card does not say, and which provider wins is not specified. Both TypeSafe AI and DigitalOcean are listed with zero data retention and no training, so if the provider matters for data-residency or contract reasons, pin it with only: ['typesafe-ai'] in providerOptions.gateway rather than relying on preference order.

Less than a typical model card: type, price, context window, provider list and a one-question snippet, plus a hosted playground. Detailed capability metadata has not been reported, and the 32,000-token window is the whole budget, so check the docs before assuming you can pass a long document as state.

The model page describes classification, routing, rubric-based assessment and automated verification. Jev evaluates shared state against typed questions and returns choices, scores and boolean probabilities rather than generated text.

Yes. Multiple questions can be evaluated in parallel within a single request against the same state, and the card's example asks one boolean question with experimental_decide. The page does not say how the shared state is billed when you add questions.

Your use is subject to TypeSafe AI's terms and privacy policies, and the provider table links Legal terms and privacy for each provider. Both TypeSafe AI and DigitalOcean are listed with zero data retention and no training flags.