← Back to Research
arxiv
LLM February 24, 2026 Impact: 9/10

Reinforcement Learning vs. Distillation: Understanding Accuracy and Capability in LLM Reasoning

By Minwu Kim, Anubhav Shrestha, Safal Shrestha, Aadim Nepal, Keith Ross
Recent studies have shown that reinforcement learning with verifiable rewards (RLVR) enhances overall accuracy (pass@1) but often fails to improve capability (pass@k) of LLMs in reasoning tasks, while distillation can improve both. In this paper, we investigate the mechanisms behind these phenomena. First, we demonstrate that RLVR struggles to improve capability as it focuses on improving the accuracy of the easier questions to the detriment of the accuracy of the most difficult questions. Second, we show that RLVR does not merely increase the success probability for the easier questions, but in our small model settings, produces quality responses that were absent in its original output distribution. Third, from the experiment distilling teacher responses to in-distribution problems, we find that capability does not always improve with distillation. We conjecture that capability improves only when new knowledge is introduced, whereas distilling reasoning patterns only improves accuracy but not capability, sacrificing performance on the most difficult questions, similar to RLVR.
reinforcement learningdistillationreasoningRLVRLLM training
Analysis

Summary

This paper tackles one of the most important open questions in LLM training methodology: when should you use reinforcement learning with verifiable rewards (RLVR) versus knowledge distillation to improve a model's reasoning capabilities? The authors draw a critical distinction between two metrics — pass@1 (accuracy: does the model get it right on the first try?) and pass@k (capability: can the model produce a correct answer among k attempts?). Their core finding is that RLVR reliably boosts pass@1 but consistently fails to improve pass@k, while distillation can improve both — but only under the right conditions.

Why It Matters

The training recipe for modern reasoning models like o1, DeepSeek-R1, and their successors heavily relies on RLVR. This paper reveals a fundamental limitation of that approach: RLVR works by optimizing the easier end of the difficulty distribution, essentially helping the model be more consistently correct on problems it already has some ability to solve. But the hardest problems — the ones that require genuine new reasoning capability — actually suffer. This is a sobering finding for anyone betting on RLVR as the primary path to superintelligent reasoning. It suggests we may hit a ceiling with pure RLVR scaling.

Distillation Is Not a Free Lunch Either

The paper's treatment of distillation is equally nuanced. The authors show that distillation only improves capability (pass@k) when it introduces genuinely new knowledge to the student model. When you distill reasoning patterns from a teacher that the student's base distribution already covers, you get accuracy gains (pass@1 improves) but no capability expansion — and you actually hurt performance on the hardest problems, mirroring RLVR's failure mode. This is a crucial insight: the source of improvement matters, not just the technique.

Implications for the Field

These findings have direct implications for how labs should think about training pipelines. If RLVR and distillation both fail to expand true capability unless novel knowledge or reasoning patterns are introduced, the field needs to invest more in curriculum design, harder problem generation, and teacher models that genuinely exceed student capability on difficult tasks. The paper also highlights a methodological gap — response length and reflection keywords are not reliable indicators of response quality — pushing for better evaluation frameworks. This is a must-read for anyone working on post-training for reasoning models.

Key Findings
1
RLVR improves pass@1 accuracy but fails to improve pass@k capability because it over-optimizes easy questions at the expense of hard ones
2
Distillation only improves capability (pass@k) when new knowledge is introduced — distilling reasoning patterns alone mimics RLVR's failure mode
3
Response length and reflection-related keywords are NOT reliable quality indicators, pointing to the need for better evaluation metrics
Impact Score
9/10
Expert Commentary
This paper delivers a rare mechanistic explanation for why RLVR hits a ceiling on hard reasoning tasks — a finding that should reshape how the field designs post-training pipelines and capability benchmarks.
D
Dong Tran
AI Researcher. Deep-dive analysis of frontier AI research papers.