TypeSafe AI ships Jev 1.14 with new box type, composable images, and longer context length
by Paul

TypeSafe AI ships Jev 1.14 with new box type, composable images, and longer context length

TypeSafe AI has released Jev 1.14, a version the company describes as deliberately unflashy: “small, subtle improvements to everything”, with the largest expected gains in multilingual support. The release arrives as jev-preview, and TypeSafe says jev-latest will roll over to 1.14 in a week, while 1.13 remains available. The company also outlined a phased deprecation schedule, promising 60 days' notice before 1.13 is formally retired and 30 days for 1.14, with a formal deprecation policy to precede both. In a notable departure from industry practice, TypeSafe said it dislikes public benchmarks but will share its internal evaluation methodology with developers anyway, framing the move as treating users “like adults”.

Alongside the incremental update, TypeSafe is opening what it calls the “danger zone”: experimental capabilities aimed at users it characterizes as dreamers rather than production builders. Chief among them is multimodal input: the company revealed that its existing Jev 1.13 model “was multimodal all along”, and image support is now exposed through a find-and-replace mechanism in assets, usable in the state, instructions, or criteria fields of its JSON interface. TypeSafe cautioned that all models, Jev included, perform worse on multimodal inputs than on pure text, and recommends decomposing questions as simply as possible. Access requires an opt-in flag: TypeSafe-Danger-Zone-Images: true.

The release also introduces a new type called box, which returns a single bounding box and is described as the company's first type not intended purely for smart-if decision statements; a preview of its broader exploration into what it calls System One Models. box is enabled only for 1.14. Separately, TypeSafe added an opt-in letting users contribute queries to help surface model “jaggedness”, while stating it will still not train on user queries. The post positions these features as a continuation of roughly two years of stealth development: less ready for automation, but an early look at what the company intends to make reliable enough for production.

Paul
Written by Paul
Follow
Add as a preferred source on Google
Jev iconJev
  17

Jev, developed by TypeSafe, is a Large Language Model (LLM) designed to enhance automation through its System One Model. It processes structured questions to deliver typed decisions, complete with probabilities and confidence levels, enabling actionable insights for software applications. Key features include an ad-free experience and structured data handling.

No comments so far, maybe you want to be first?