# TypeSafe AI releases Jev 1.14

The update introduces experimental API flags for image inputs, bounding box outputs, a 4x context length increase, and multilingual improvements.

- Image inputs require the TypeSafe-Danger-Zone-Images header and can be placed anywhere in the JSON interface on versions 1.13.0 and 1.14.0
- A machine-native box output type returns a single bounding box without type errors exclusively on version 1.14.0
- Extended context limits can be unlocked using the TypeSafe-Danger-Zone-Long-Context header
- The jev-preview endpoint now routes to jev-1.14.0, and jev-latest will point to it in one week
- Version 1.13.0 will receive 60 days of deprecation notice before removal

## Why it matters

Developers using TypeSafe's Jev models can now test multimodal inputs, extract bounding boxes, and process larger context windows using opt-in API headers.

## Sources

- [TypeSafe AI: TypeSafe AI releases Jev 1.14 model update](https://x.com/typesafeai/status/2110536685805355064)

---

Summarized by dstilled on 2026-10-14. https://dstilled.ai/story/28faac0e-efd5-421c-b773-a54027803503