“Technical model review, read the official paper by ByteDance.”




For the first time in history, a model’s core intelligence is no longer static. Vision CLv1 features a unique breakthrough where the main model weights themselves are updated during inference.
This means that for every individual and every instance, a uniquely personalized version of the model is created on the fly. It doesn't just "process" your input—it adapts its entire neural state to provide a truly unique, human-like reasoning experience.
We envision a future where AGI is completely decentralized. We are not just building software models; we are an AGI-driven hardware company actively researching and developing custom Neural Processing Units (NPUs) designed specifically for local inference of our best AI models. Removing the dependency on centralized servers is what true AGI means to us.
A dramatic, chronological breakdown of our model versions.
Step inside the architecture that redefined the AGI race. Experience Vision CLv1, a model that doesn’t just process data, but learns, adapts, and gains experience in real-time. The era of static intelligence is officially over.