TypeSafe releases Jev 1.14 with experimental features
TypeSafe has shipped Jev 1.14.0 and is putting image input, a bounding-box output type and longer context behind opt-in "danger zone" API flags, splitting features it calls production-ready from those it doesn't.
TypeSafe announced Jev 1.14.0, a modest update to its model, along with three experimental features that developers turn on with "danger zone" API headers. The company says these features will carry over to future models but are still too "jagged" for it to recommend in production.
Images. TypeSafe says the current model, jev-1.13.0, "was multimodal all along." Images are inserted through find-and-replace in assets and can go anywhere the JSON interface allows: state, instructions or criteria. The company warns that all models, Jev included, do worse on images than on plain text. The header is TypeSafe-Danger-Zone-Images: true.
Box type. A new box type returns a single bounding box. TypeSafe calls it its first type that isn't purely meant for "smart-if" decisions. It works only on jev-1.14.0.
Long context. Jev supports more context than TypeSafe has allowed in production. The company says it set the lower limit on purpose because reliability drops on long inputs. Users can now lift it with TypeSafe-Danger-Zone-Long-Context: true. The post does not say how long the full context is.
Jev 1.14.0 is available under jev-preview. TypeSafe describes it as small improvements across the board, with the biggest gain expected in multilingual support. jev-latest will move to 1.14.0 in a week. TypeSafe plans to give 60 days' notice before deprecating 1.13.0 and 30 days for 1.14.0, and says a formal deprecation policy will come first.
Users can also opt in to let TypeSafe use their queries to find model weaknesses. The company says it still won't train on queries.
The post also mentions internal intelligence measurements compared only against open-weight models. TypeSafe says the numbers may be biased in its favor, but no figures appear in the text.