TypeSafe AI Releases Jev 1.14 With Images, Box Type and 262k Context
MarkTechPost Asif Razzaq
TypeSafe AI shipped Jev 1.14 with image input, a box type, and a bigger context mode. All three are marked experimental, so this is more preview than product.
Based on reporting by MarkTechPost, Asif Razzaq — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
TypeSafe AI has pushed out Jev 1.14, the newest version of its System One decision model, and it is already live under the jev-preview alias. The headline features are image support, a new Box question type, and a longer context mode. None of those is considered ready for production yet.
Jev is not a chatty generator. It returns typed decisions instead of free-form text, taking a state plus questions such as Choice, Score or Noul and replying with typed values, probabilities and a confidence score. TypeSafe says the 1.14.0 release brings small improvements across the board, with the biggest gains expected in multilingual support.
The image side is intentionally fussy. Images go into a top-level assets map, get referenced in state or instructions, and are swapped in before the model reads them. TypeSafe says Jev 1.13 already supported images, but both versions require the TypeSafe-Danger-Zone-Images: true header. The limits are narrow: PNG or JPEG only, up to 16 MiB per file and 32 MiB per request body. The company also warns that image work is weaker than pure text, so it tells developers to keep the questions simple.
Box is the other new trick. It returns one bounding box per question, with top, bottom, left and right edges given as fractions from 0 to 1. But there is a catch: Box always returns a box, even when the object is missing, so TypeSafe recommends pairing it with a Noul question that checks whether anything is actually there. In a small test on 36 painting crops, four per-edge Noul questions caught 30 cut-off crops, while a single “is all of it inside” question caught 26.
The context window got a similar treatment. By default, Jev keeps 64k tokens per request and 32k per question. An opt-in header raises both limits to 262k, but TypeSafe says it kept context capped on purpose because long inputs are less reliable. Pricing stays at $42 per billion input tokens, or $0.042 per million, with output tokens still free. jev-latest moves to 1.14.0 in a week, while 1.13.0 remains available for now.
My take — AI-written commentary, not fact-checked reporting
This is a very TypeSafe move: ship the sharp edges first, label them danger-zone, then ask developers to do the sensible thing anyway. The company is basically admitting that bigger context and vision support are useful, but not magic, which is a healthier pitch than the usual AI confetti cannon. More teams should say that out loud instead of pretending every preview is production with a nicer jacket.
Read more about this at: MarkTechPost
