AI HOT (Curated Pool)Jun 24, 2026, 11:32 UTC+8

Qwen-AgentWorld open-sourced: an agent that predicts before it acts

Qwen-AgentWorld 开源：让 Agent 学会"先预测，再行动"

Qwen released Qwen-AgentWorld, a native language world model covering seven domains: MCP, Search, Terminal, SWE, Web, OS, and Android. Trained on over 10 million real interaction trajectories through CPT→SFT→RL, it scored 58.71 on AgentWorldBench, edging out GPT-5.4 (58.25) and Claude Opus 4.8. As a decoupled environment simulator, it hit 50.3% F1 on WideSearch via Sim RL, beating real-environment RL at 45.6%. When used as an agent foundation model with LWM warm-up, it transfers to seven benchmarks—three of which never appeared in training. Both model and benchmark are open-sourced.

Why it matters: Qwen dropped an agent model with a clear methodology and benchmark — not concept hype. The 7-environment coverage and 10M+ training traces make it substantive, but it just went open-source and the community hasn't reproduced it yet, so the score stays below 85.

Read the original ↗Export Markdown

Take

The reason to click: Qwen trained an agent that predicts what the environment will look like before taking an action, then open-sourced both the model and the benchmark.

They call it a "native language world model." The concrete part: it covers seven domains—MCP, search, terminal, software engineering, web, OS, and Android—trained on over 10 million real interaction trajectories through a CPT→SFT→RL pipeline.

On their in-house AgentWorldBench, the 397B-total MoE model scored 58.71, just ahead of GPT-5.4 at 58.25 and Claude Opus 4.8. The margin is small, but the open-source angle matters more. You can use it as a decoupled environment simulator for Sim RL—on WideSearch it hit 50.3% F1, beating real-environment RL at 45.6%. Or use it as an agent foundation model: with LWM warm-up, it transfers to seven benchmarks, three of which never appeared in training. That transfer is more interesting than the raw score.

Where I'd discount: the benchmark is their own, no external validation yet, and a 397B-parameter MoE isn't cheap to run. But the open release makes this worth tracking.

Source: AI HOT (Curated Pool) · mp.weixin.qq.com
Published: Jun 24, 2026, 11:32 UTC+8 · Jun 24, 2026