**[TypeSafe Upgrades Jev 1.14: Can Locate Objects in Images, Returns Coordinates Even if Not Found]** Beating AI Newsflash: TypeSafe AI has released Jev 1.14, introducing experimental visual capabilities to this model designed for rapid decision-making. Developers can use it to identify whether there are people in a camera feed or compare two photos to determine if they depict the same dog. The newly added Box localization feature can pinpoint the location of specified objects in an image, returning the coordinates of a rectangular region to guide other programs on where to crop or operate. TypeSafe CEO Diogo Almeida revealed that the previous version, Jev 1.13, was already capable of processing images, but the team refrained from enabling visual tasks due to concerns about accuracy reliability. Now, both versions support image inputs, allowing pictures to be directly used as conditions or options for decision-making. However, Box localization is exclusive to version 1.14. In experimental mode, the context limit for a single request has been expanded from 64,000 tokens to approximately 262,000 tokens—four times the original capacity. The new version also improves overall judgment capabilities, with the most noticeable enhancement expected in multilingual support, though specific details on the improvement have not been disclosed. Jev 1.14 is currently available via `jev-preview` and is expected to become the default stable version on October 21. However, TypeSafe has explicitly cautioned that both the visual and extended context features remain in the experimental phase and are not yet production-grade. Extended context functionality may increase latency and potentially reduce accuracy. Additionally, Box localization has a unique limitation: even if the specified object is absent in the image, it will still return a coordinate box. Developers must therefore verify the presence of the target object to prevent subsequent programs from operating on incorrect locations. [Original Link]