Zing-0.5 Ships 5B Playable-World Model at 24 FPS With Joint Keyboard and Text Control for ~$0.009 per Stream-Minute

Zing-0.5 is a 5B autoregressive world model that lets users navigate generated environments with keyboard input while injecting event-shaping text commands in real time. Using distribution-matching distillation from a multi-prompt teacher, it runs at 24 FPS at 832x480 for roughly $0.009 per stream-minute and scores 81.0 overall and 88.5 consistency on 158 WBench Navigation cases. Weights, inference code, and serving stack are on Hugging Face and GitHub.