IT Home Releases Three LingBot-World 2.0 AI Models, One for Consumer GPUs
Chinese tech entity IT Home has open-sourced three new LingBot-World 2.0 AI models. The Small version, with 1.3 billion parameters, is designed for single consumer-grade GPUs.
IT Home, a Chinese technology entity, has released three new AI models under the LingBot-World 2.0 series as open-source software. Following the 14 billion parameter model released in July, the company has now introduced LingBot-World 2.0 Small (1.3B), LingBot-World 2.0 Bidirectional, and LingBot-World 2.0 Causal Pretrain.
The LingBot-World 2.0 series is designed for real-time interactive environments. These models can generate content with hourly continuity and respond to character actions such as attacking or shooting, as well as environmental events like snowfall. With higher computing power and corresponding inference configurations, it can achieve a 720p resolution at 60 FPS.
The LingBot-World 2.0 Small model, specifically, features a 1.3 billion parameter scale optimized for consumer-grade GPUs, enabling real-time generation on a single card. IT Home stated this version is intended for individual developers, university labs, and small teams to explore interaction methods, inference optimization, and application scenarios.
The LingBot-World 2.0 Bidirectional model is intended for training real-time world models and provides a high-quality distillation source. The LingBot-World 2.0 Causal Pretrain model is designed for post-training, evaluation, and scenario adaptation, serving as a causal pre-training base.
The models are available for download on platforms including Hugging Face and ModelScope. IT Home has also published a technical report detailing the models.