Qwen-AgentWorld Officially Launched by Qianwen

On June 24, Qianwen officially launched Qwen-AgentWorld—the first native Language World Model (LWM) capable of simulating agent interaction environments across seven major domains. Native world modeling: Environmental modeling has been the training objective since the Continued Pre-Training (CPT) phase, spanning the entire process from CPT to Supervised Fine-Tuning (SFT) to Reinforcement Learning (RL), rather than being an after-the-fact adaptation of a general large language model. One model for seven domains: A single model covers both text-based environments (MCP, Search, Terminal, SWE) and GUI-based environments (Web, OS, Android), enabling cross-domain knowledge transfer.