---
title: "Qwen's AgentWorld-35B Beats GPT-5.4 by Simulating Environments Instead of Acting in Them"
subtitle: "Qwen releases the first open language world model that can simulate 7 agent environments and directly improve downstream agent performance across benchmarks it was never trained on."
url: "https://alphasignal.ai/news/qwen-s-agentworld-35b-beats-gpt-5-4-by-simulating-environments-instead-of"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "Unsloth AI"
company_url: "https://alphasignal.ai/company/unslothai"
published: "2026-06-25T05:51:07.000Z"
updated: "2026-07-02T02:52:17.979Z"
type: "Model"
topics:
  - "Agents"
  - "Llms"
subtopics:
  - "Tool Use"
  - "Multi Agent"
upvotes: 3562
source_url: "https://huggingface.co/unsloth/Qwen-AgentWorld-35B-A3B-GGUF"
comments: 0
truncated: true
---

# Qwen's AgentWorld-35B Beats GPT-5.4 by Simulating Environments Instead of Acting in Them

> Qwen releases the first open language world model that can simulate 7 agent environments and directly improve downstream agent performance across benchmarks it was never trained on.

Training AI agents has always required real environments: live terminals, actual browsers, running sandboxes. Qwen just proposed a different path. [Qwen-AgentWorld](https://huggingface.co/Qwen/Qwen-AgentWorld-35B-A3B) is a language model trained not to act in environments, but to **simulate them** , predicting what the environment returns after any given action. The 35B-A3B variant is now available as a quantized GGUF from Unsloth, making it runnable on consumer hardware.

## The gap nobody was filling

General agents require robust world modeling capability: learning to predict environment state transitions is both necessary and foundational for effective action selection. Prior LLM agent research disproportionately emphasized state-to-action policy learning, neglecting comprehensive modeling of environment dynamics. Qwen-AgentWorld directly addresses that gap.

The result is the first language world model that simulates seven agent environments through long chain-of-thought reasoning: MCP, Search, Terminal, Software Engineering, Android, Web, and OS. For the three GUI domains, environment observations are represented as accessibility trees and UI view hierarchies rather than pixel frames.

## How it was built

Leveraging more than 10M environment interaction trajectories across 7 domains in real-world environments, Qwen-AgentWorld was developed through a three-stage training pipeline: CPT injects general-purpose world modeling capabilities from state transition dynamics and augmented professional corpora, SFT activates next-state-prediction reasoning, and RL sharpens simulation fidelity through a tailored framework with hybrid rubric-and-rule rewards.

The model's architecture is a sparse Mixture-of-Experts (MoE) design , 35B total parameters but only 3B are active during inference, keeping compute costs manageable. It also uses a hybrid attention mechanism combining Gated DeltaNet linear attention layers with standard Gated Attention, and supports a 262,144-token context window, which is critical for long multi-turn environment trajectories.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [unsloth/Qwen-AgentWorld-35B-A3B-GGUF · Hugging Face](https://huggingface.co/unsloth/Qwen-AgentWorld-35B-A3B-GGUF)
- [Qwen-AgentWorld: Language World Models for General Agents (arXiv)](https://arxiv.org/abs/2606.24597)
- [QwenLM/Qwen-AgentWorld GitHub Repository](https://github.com/QwenLM/Qwen-AgentWorld)
- [Qwen-AgentWorld Blog Post](https://qwen.ai/blog?id=qwen-agentworld)
- [Qwen/AgentWorldBench Dataset](https://huggingface.co/datasets/Qwen/AgentWorldBench)