Jev 1.14

Jev 1.14 — addressable as jev-1.14.0, or via the jev-preview alias — is the TypeSafe AI System One version that opened the vision path. It adds the box question type, accepts image input, and lifts the long-context ceiling, while every existing text decision keeps working unchanged. Here's what changed from 1.13, how to pin the version, and whether you should.

Jev versions are pinned checkpoints of TypeSafe AI's System One model. jev-latest tracks the current stable release (jev-1.13.0); jev-1.14.0 is the newer line that adds vision and is also reachable as jev-preview. You choose the version with the model field on a decide call — we forward whatever you set and never override it — so upgrading is a one-word change, and so is pinning for stability.

What changed in 1.14

• box — a new question type that locates a described target in an image as 0–1 fractional coordinates (requires 1.14)
• Images — a decide call can carry pictures in a top-level assets map (images also work on 1.13)
• Long context — one header lifts both input budgets to ~262k tokens, at the same per-token price
• jev-preview alias — jev-preview points at jev-1.14.0, so you can try the newest line without pinning the exact number
• Unchanged — existing choice / score / noul text calls behave exactly as before

jev-1.14.0 vs jev-1.13.0

How to pin the version

Set the model field explicitly. Omit it and you get jev-latest (1.13.0, the stable line). Pin jev-1.14.0 (or jev-preview) when you need box or want to test the newest line. The vision features are preview — the field and header names are TypeSafe's own and the exact shapes may still move — so pin the exact version rather than jev-latest if your code depends on them.

// Pin the version with the model field — we never override it.
const body = {
  model: "jev-1.14.0",          // or "jev-preview"; omit for jev-latest (1.13.0)
  state: { photo: "shelf" },
  assets: { shelf: { type: "image", data_base64: imgB64, format: "png" } },
  questions: {
    present: { type: "noul", instructions: "Is there a red mug in `photo`?" },
    where:   { type: "box",  instructions: "the red mug in `photo`" }, // box needs 1.14
  },
};
// Vision requests opt in with one header:
//   TypeSafe-Danger-Zone-Images: true   (forwarded verbatim)

Should you upgrade?

If you only make text decisions, there's nothing to do — 1.13.0 is stable and your calls are unchanged. Reach for 1.14 when you want the vision path (images or the box question) or the higher long-context ceiling. Because those are preview, pin jev-1.14.0 explicitly in production so a later preview change can't shift behaviour under you. The full vision how-to — sending images, the box answer shape, limits and cost — is on the multimodal Jev page.

FAQ

See also: Multimodal Jev — images & the Box question · Jev context window & long context · Jev models & versions · Docs — Vision & advanced · Pricing