![[2508.05004] R-Zero: Self-Evolving Reasoning LLM from Zero Data](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2508.05004] R-Zero: Self-Evolving Reasoning LLM from Zero Data
Summary
The article presents R-Zero, a self-evolving reasoning LLM that autonomously generates training data, improving AI capabilities without human intervention.
Why It Matters
R-Zero addresses the limitations of traditional LLM training methods that rely on human-curated data. By enabling models to learn from their own experiences, it paves the way for more advanced AI systems, potentially leading to super-intelligence. This innovation could significantly enhance reasoning abilities in AI, making it a crucial development in the field of machine learning.
Key Takeaways
- R-Zero creates its own training data from scratch, eliminating reliance on pre-existing tasks.
- The model consists of two roles: Challenger and Solver, which co-evolve through interaction.
- Empirical results show significant improvements in reasoning capabilities across various benchmarks.
- This framework could lead to advancements in AI beyond human intelligence.
- R-Zero represents a shift towards more autonomous AI training methodologies.
Computer Science > Machine Learning arXiv:2508.05004 (cs) [Submitted on 7 Aug 2025 (v1), last revised 13 Feb 2026 (this version, v4)] Title:R-Zero: Self-Evolving Reasoning LLM from Zero Data Authors:Chengsong Huang, Wenhao Yu, Xiaoyang Wang, Hongming Zhang, Zongxia Li, Ruosen Li, Jiaxin Huang, Haitao Mi, Dong Yu View a PDF of the paper titled R-Zero: Self-Evolving Reasoning LLM from Zero Data, by Chengsong Huang and 8 other authors View PDF HTML (experimental) Abstract:Self-evolving Large Language Models (LLMs) offer a scalable path toward super-intelligence by autonomously generating, refining, and learning from their own experiences. However, existing methods for training such models still rely heavily on vast human-curated tasks and labels, typically via fine-tuning or reinforcement learning, which poses a fundamental bottleneck to advancing AI systems toward capabilities beyond human intelligence. To overcome this limitation, we introduce R-Zero, a fully autonomous framework that generates its own training data from scratch. Starting from a single base LLM, R-Zero initializes two independent models with distinct roles, a Challenger and a Solver. These models are optimized separately and co-evolve through interaction: the Challenger is rewarded for proposing tasks near the edge of the Solver capability, and the Solver is rewarded for solving increasingly challenging tasks posed by the Challenger. This process yields a targeted, self-improving curriculum without any pre-...