TypeSafe AI released Jev 1.14.0, which the company describes as an unflashy, small update with broad but subtle changes and the clearest expected gains in multilingual performance; it warns that non-English languages, Chinese included, still lag English. The release also opens three experimental features grouped under a "Danger Zone" program: image input, a new Box question type that returns normalized bounding-box coordinates, and long context that raises per-request and per-question limits to 262k tokens. All three are off by default and require manual API switches, and TypeSafe says they do not meet production standards and are used at the developer's own risk. Pricing is unchanged at $0.042 per million input tokens with free output, and TypeSafe's internal evaluation put Jev 1.14.0 first among eight models at 60.0 while cautioning the benchmark may favor its own model and scores may change. DoNews also reported that Liquid AI launched its d1-ultra decision model the next day, claiming it beats Jev 1.14 across eight task types, a vendor claim.