Jev 1.14 stops writing prose: one request returns typed, probabilistic answers for routing, scoring and checks
Synopsis
TypeSafe AI's decision model Jev 1.14 is now available on AI Gateway. It takes a state and a set of questions and returns typed Choice, Score and Boolean answers with probabilities instead of free-form text, answering every question in one request in parallel for classification, routing, scoring and automated verification. Version 1.14 accepts images as part of the state and adds an experimental long-context mode for documents, logs or conversation histories too long for a standard request.
Interpretation
Software decision tasks can now ask a model for typed answers with probabilities rather than a block of free-form text. The earlier pattern hands state to a model and reads back natural-language output that outside code must parse into usable fields. Jev 1.14 is offered on AI Gateway under the model name typesafe-ai/jev-1.14.0; callers pass state and a question set and receive Choice, Score and Boolean answers.
All questions are answered in parallel within a single request, so classification, routing, scoring and automated verification can share one call. AI Gateway documentation describes decision models as a separate interface from free-form text generation. The example writes questions as typed entries with instructions, such as a Boolean question asking whether an image is predominantly red, and reads results from the answers field by question name.
Images can be passed as part of the state, and long documents go through an experimental long-context mode. Documents, logs or conversation histories beyond a standard request were not part of the regular input scope. An image enters state as a file part alongside text; long context is enabled with providerOptions.typesafe.longContext, and the example reads a support log to check whether it reports a damaged delivery.
Perspective
The report covers Jev 1.14's availability on AI Gateway, its input form and return types, for developers who need structured decision results inside software workflows. Image input uses a file part, and long context is switched on with providerOptions.typesafe.longContext; long context is explicitly labeled experimental. The source states no follow-up evaluation or migration plan.
Long-context mode is labeled experimental, and the source does not state where its applicability ends. The source also gives no accuracy, latency or comparison against free-text-plus-parsing approaches, so the reliability of typed answers in real systems remains for users to verify.