Tongyi Qianwen launches the native language world model Qwen-AgentWorld

Alitongyi officially launched Qwen-AgentWorld, the first native language world model, which can simulate intelligent agent interaction environments in seven major domains: MCP, Search, Terminal, SWE, Web, OS, and Android. The model is based on over 10 million real-world environmental trajectories and undergoes end-to-end training in three stages: CPT→SFT→RL. On the self-developed benchmark AgentWorldBench, its overall simulation quality surpasses top models such as GPT-5.4 and Claude Opus 4.8.