TypeSafe AI Ships Jev 1.14.0 With “Danger Zone” Images, Bounding Boxes, and 262K Context
TypeSafe’s first major update since launch upgrades multilingual performance in Jev 1.14.0 and unlocks opt-in headers for composable images, a new box type, and 262K context.
One month after debuting Jev, TypeSafe AI shipped its first major model update on October 14—introducing jev-1.14.0 alongside a bold new “Danger Zone” header mechanism for experimental capabilities that haven’t yet cleared TypeSafe’s strict production reliability bar.

Jev 1.14.0 unlocks composable images, typed bounding boxes, and 262K context behind explicit Danger Zone opt-in headers.
Multimodal all along, plus a new box primitive
In a playful blog post titled “Jev: For the builders AND dreamers,” TypeSafe revealed that even its original launch model (jev-1.13.0) had been multimodal under the hood. Sending TypeSafe-Danger-Zone-Images: true unlocks composable PNG and JPEG image assets anywhere inside JSON state, instructions, or criteria. Exclusively on jev-1.14.0, TypeSafe also introduced box—a typed bounding-box output with zero schema errors.
curl https://api.typesafe.ai/v1/systemone \
-H "Authorization: Bearer $TYPESAFE_API_KEY" \
-H "TypeSafe-Danger-Zone-Images: true" \
-H "TypeSafe-Danger-Zone-Long-Context: true" \
-d '{"model":"jev-1.14.0","state":{"screen":"<image>"},"questions":{"target":{"type":"box","instructions":"Locate the checkout button"}}}'
Deprecation & privacy commitments
- 60 days’ warning promised before formally deprecating
jev-1.13.0(and 30 days forjev-1.14.0). - 262K long context stays at $0.042 per million input tokens, with TypeSafe warning that longer contexts trade away speed and reliability.
- Opt-in debugging: developers can flag jagged queries for TypeSafe engineers without allowing training on customer data.