Paper 2603.08660
How Far Can Unsupervised RLVR Scale LLM Training?
- Published
- Mar 2026
- Research lab
- Independent
- Citations
- 24
- GitHub
- 1.1K stars
01 In brief
Summary
This paper investigates the scalability of Unsupervised Reinforcement Learning with Verifiable Rewards (URLVR) for LLM training.
It categorizes URLVR methods into intrinsic (certainty-based and ensemble-based) and external rewards.
The authors establish a unified theoretical framework showing that all intrinsic methods converge to sharpening the model's initial distribution, amplifying existing preferences.
This leads to a consistent rise-then-fall performance pattern across methods, with collapse timing determined by model prior.
Intrinsic rewards are safe for test-time training on small datasets, and the proposed Model Collapse Step serves as an efficient indicator of RL trainability.
External rewards, such as self-verification, exploit generation-verification asymmetries and show sustained improvement without collapse, suggesting a more scalable path.
02 From the paper
Abstract
Unsupervised reinforcement learning with verifiable rewards (URLVR) offers a pathway to scale LLM training beyond the supervision bottleneck by deriving rewards without ground truth labels. Recent works leverage model intrinsic signals, showing promising early gains, yet their potential and limitations remain unclear. In this work, we revisit URLVR and provide a comprehensive analysis spanning taxonomy, theory and extensive experiments. We first classify URLVR methods into intrinsic versus external based on reward sources, then establish a unified theoretical framework revealing that all intrinsic methods converge toward sharpening the model's initial distribution This sharpening mechanism succeeds when initial confidence aligns with correctness but fails catastrophically when misaligned. Through systematic experiments, we show intrinsic rewards consistently follow a rise-then-fall pattern across methods, with collapse timing determined by model prior rather than engineering choices. Despite these scaling limits, we find intrinsic rewards remain valuable in test-time training on small datasets, and propose Model Collapse Step to measure model prior, serving as a practical indicator for RL trainability. Finally, we explore external reward methods that ground verification in computational asymmetries, showing preliminary evidence they may escape the confidence-correctness ceiling. Our findings chart boundaries for intrinsic URLVR while motivating paths toward scalable alternatives.