Nex-N1: Agentic Models Trained via a Unified Ecosystem for Large-Scale Environment Construction
Nex-AGI Team, Yuxuan Cai, Lu Chen, Qiaoling Chen, Yuyang Ding, Liwen Fan, Wenjie Fu, Yufei Gao, Honglin Guo, Pinxue Guo, Zhenhua Han, Zhengfu He, Hanglei Hu, Kai Hu, Shengjia Hua, Tianyu Huai, Baodai Huang, Li Ji, Zhen Jiang, Zhikai Lei, Bufan Li, Jiahang Lin
2025-12-05
Summary
This paper focuses on making Large Language Models, or LLMs, more capable of acting as independent agents that can make decisions and complete complex tasks, rather than just responding to prompts.
What's the problem?
Currently, LLMs are good at mimicking what they've been taught, but they struggle when faced with situations requiring real-world problem-solving and independent action. A major roadblock to improving this is the difficulty in creating enough realistic and varied practice environments to effectively 'train' these agents to make good decisions. It's hard to build systems that can automatically generate the kinds of complex interactions needed for this training.
What's the solution?
The researchers developed a system called Nex, which includes three main parts. First, NexAU allows for building complicated agent setups easily. Second, NexA4A automatically creates a wide range of these agent setups from simple descriptions. Finally, NexGAP connects these simulated environments to the real world, making the training more practical. They then used this system to train a new LLM, Nex-N1, in these diverse environments.
Why it matters?
This work is important because it provides a way to reliably create the training grounds needed for LLMs to become truly autonomous agents. Nex-N1 performed very well on challenging tasks, even competing with the best existing models, and the researchers are sharing their tools and model so others can build upon this research and further advance the field of AI.
Abstract
The evolution of Large Language Models (LLMs) from passive responders to autonomous agents necessitates a fundamental shift in learning paradigms -- from static imitation to incentive-driven decision making. However, this transition is significantly impeded by the lack of scalable infrastructure capable of constructing high-quality interaction signals for effective policy learning. To address this, we introduce a comprehensive method designed to systematically scale the diversity and complexity of interactive environments. Our method realizes this scaling by addressing three orthogonal dimensions: (1) Complexity: NexAU, a flexible agent framework that supports building complex agent hierarchies via simple configurations; (2) Diversity: NexA4A automatically generates diverse agent hierarchies from natural language to cover infinite domains; and (3) Fidelity: NexGAP bridges the simulation-reality gap by integrating dynamic real-world environment for grounded trajectories synthesis. We train Nex-N1 upon the diverse and complex interactive environments established by our infrastructure. Empirical results on benchmarks such as SWE-bench and tau2 demonstrate that Nex-N1 consistently outperforms SOTA open-source models and achieves competitive performance against frontier proprietary models on complex agentic tasks. We open-source the Nex ecosystem and model weights to facilitate further research.