88Z AI, 24 June — Alibaba Tongyi has launched Qwen-AgentWorld, the first native language world model, supporting agent interaction simulation across seven domains. Trained on tens of millions of real-world trajectories, it outperforms top models like GPT-5.4 and Claude Opus 4.8 in proprietary benchmarks.
At a glance
- Qwen-AgentWorld is the first native language world model covering seven interaction domains
- Trained on tens of millions of real trajectories, its simulation quality surpasses top models like GPT-5.4
The update
Alibaba Tongyi has officially released Qwen-AgentWorld, the industry's first native language world model. This model can accurately simulate agent interaction environments across seven core domains: MCP, Search, Terminal, Software Engineering, Web, OS, and Android. It is trained on over ten million real-world environment trajectories using a three-stage end-to-end process comprising continuous pre-training, supervised fine-tuning, and reinforcement learning. In the proprietary AgentWorldBench benchmark, Qwen-AgentWorld demonstrated superior overall simulation quality, outperforming several leading models including GPT-5.4 and Claude Opus 4.8, showcasing robust capabilities in environment understanding and interaction simulation.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 千问大模型Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.