Conference paper · 2025 · Open access
Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?
neural information processing systems, 64304–64339 · Published 1 January 2025 · DOI 10.52202/085713-1933
In brief
“Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?” is a 2025 conference paper by Zhiqi Chen and 6 co-authors, published in neural information processing systems. OpenAlex counts 821 citations as of 15 November 2026. It is open access under a CC BY licence: the full text is free to read at arXiv (Cornell University).
- Citations
- 821
- Most cited in
- 2026 (516)
- Field-weighted impact
- 289.98
- References
- 30
- Authors
- 7
Abstract
Reinforcement Learning with Verifiable Rewards (RLVR) has recently demonstrated notable success in enhancing the reasoning performance of large language models (LLMs), particularly on mathematics and programming tasks. Similar to how traditional RL helps agents explore and learn new strategies, RLVR is believed to enable LLMs to continuously self-improve, thus acquiring novel reasoning abilities beyond those of the corresponding base models. In this study we critically examine the current state of RLVR by systematically probing the reasoning capability boundaries of RLVR-trained LLMs across various model families, RL algorithms, and math, coding, and visual reasoning benchmarks, using pass@k at large k values as the evaluation metric. Surprisingly, we find that the current training setup does not elicit fundamentally new reasoning patterns. While RLVR-trained models outperform their base models at small k (e.g., k = 1), the base models achieve a higher pass@k score when k is large. Coverage and perplexity analyses show that the observed reasoning abilities originate from and are bounded by the base model. Treating the base model as an upper bound, our quantitative analysis shows that six popular RLVR algorithms perform similarly and remain far from optimal in leveraging the potential of the base model. By contrast, we find that distillation can introduce new reasoning patterns from the teacher and genuinely expand the model's reasoning capabilities. Overall, our findings suggest that current RLVR methods have not yet realized the potential of RL to elicit truly novel reasoning abilities in LLMs. This highlights the need for improved RL paradigms, such as continual scaling and multi-turn agent-environment interaction, to unlock this potential.
Cite this paper
Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.
APA 7
Chen, Z., Lu, R., Zhao, A., Wang, Z., Yue, Y., Song, S., & Huang, G. (2025). Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? neural information processing systems, 64304–64339. https://doi.org/10.52202/085713-1933
MLA 9
Chen, Zhiqi, et al. “Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?” neural information processing systems, 2025, pp. 64304–64339. https://doi.org/10.52202/085713-1933.
Chicago
Chen, Zhiqi, Rui Lu, Andrew Zhao, Zhaokai Wang, Yang Yue, Shiji Song, and Gao Huang. 2025. “Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?” neural information processing systems: 64304–64339. https://doi.org/10.52202/085713-1933.
Harvard
Chen, Z. et al. (2025) ‘Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?’, neural information processing systems, pp. 64304–64339. doi:10.52202/085713-1933.
IEEE
Z. Chen et al., “Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?,” neural information processing systems, pp. 64304–64339, 2025, doi: 10.52202/085713-1933.
BibTeX
@inproceedings{chen2025does,
title = {{Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?}},
author = {Chen, Zhiqi and Lu, Rui and Zhao, Andrew and Wang, Zhaokai and Yue, Yang and Song, Shiji and Huang, Gao},
booktitle = {neural information processing systems},
year = {2025},
pages = {64304--64339},
doi = {10.52202/085713-1933},
url = {https://doi.org/10.52202/085713-1933}
}
Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)
Citations by year
| Year | Citations | Share of the best year |
|---|---|---|
| 2024 | 1 | |
| 2025 | 304 | |
| 2026 | 516 |
Its field-weighted citation impact is 289.98: it has been cited 289.98 times as often as the average work of the same field, type and year (1.00 is the world average).
Topic and keywords
Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- reinforcement learning with verifiable rewards
- large language models
- coverage analysis
- knowledge distillation
- mathematical reasoning
- visual reasoning
Work with this paper
Most-cited papers in Computer Science
- ImageNet classification with deep convolutional neural networksAlex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton · Communications of the ACM · 2017 · 114,726 citations
- Long Short-Term MemorySepp Hochreiter, Jürgen Schmidhuber · Neural Computation · 1997 · 109,097 citations
- A Mathematical Theory of CommunicationClaude E. Shannon · Bell System Technical Journal · 1948 · 91,736 citations
- Deep learningYann LeCun, Yoshua Bengio, Geoffrey E. Hinton · Nature · 2015 · 90,226 citations
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksShaoqing Ren, Kaiming He, Ross Girshick and 1 more · IEEE Transactions on Pattern Analysis and Machine Intelligence · 2016 · 74,143 citations
Questions about this paper
How do I cite this paper in APA style?
Chen, Z., Lu, R., Zhao, A., Wang, Z., Yue, Y., Song, S., & Huang, G. (2025). Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? neural information processing systems, 64304–64339. https://doi.org/10.52202/085713-1933
How many times has this paper been cited?
821 times, according to OpenAlex on 15 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.
Is this paper free to read?
Yes. It is open access: read the free full text at arXiv (Cornell University).
Who wrote this paper?
Zhiqi Chen, Rui Lu, Andrew Zhao, Zhaokai Wang, Yang Yue, Shiji Song, Gao Huang.
What is the DOI of this paper?
10.52202/085713-1933. The permanent link is https://doi.org/10.52202/085713-1933.
Bibliographic data from OpenAlex (CC0), last checked 15 November 2026. ResearchGuru does not host this paper; the links go to the publisher or the repository that does.