Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?

• URL: http://arxiv.org/abs/2510.10541v1
• Date: Sun, 12 Oct 2025 10:49:57 GMT
• Title: Rethinking RL Evaluation: Can Benchmarks Truly Reveal Failures of RL Methods?
• Abstract summary: We find that training on benchmarks achieves nearly the same performance as training directly on the test sets.<n>We propose three core principles for designing more faithful benchmarks: sufficient difficulty, balanced evaluation, and distributional robustness.
• Score: 43.02103699297655
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Current benchmarks are inadequate for evaluating progress in reinforcement learning (RL) for large language models (LLMs).Despite recent benchmark gains reported for RL, we find that training on these benchmarks' training sets achieves nearly the same performance as training directly on the test sets, suggesting that the benchmarks cannot reliably separate further progress.To study this phenomenon, we introduce a diagnostic suite and the Oracle Performance Gap (OPG) metric that quantifies the performance difference between training on the train split versus the test split of a benchmark. We further analyze this phenomenon with stress tests and find that, despite strong benchmark scores, existing RL methods struggle to generalize across distribution shifts, varying levels of difficulty, and counterfactual scenarios: shortcomings that current benchmarks fail to reveal.We conclude that current benchmarks are insufficient for evaluating generalization and propose three core principles for designing more faithful benchmarks: sufficient difficulty, balanced evaluation, and distributional robustness.

Related papers

• What If Consensus Lies? Selective-Complementary Reinforcement Learning at Test Time [57.533031432715084]
  Test-Time Reinforcement Learning (TTRL) enables Large Language Models (LLMs) to enhance reasoning capabilities on unlabeled test streams.<n>Existing TTRL methods rely exclusively on positive pseudo-labeling strategies.<n>We propose SCRL, a robust test-time reinforcement learning framework that effectively mitigates label noise amplification.
  arXiv  Detail & Related papers  (2026-03-20T11:47:12Z)
• Test-time RL alignment exposes task familiarity artifacts in LLM benchmarks [20.988724384459392]
  The train-before-test approach controls for task familiarity by giving each model task-relevant training before evaluation.<n>We propose a two-stage test-time reinforcement learning (RL) alignment method for train-before-test.
  arXiv  Detail & Related papers  (2026-03-13T10:24:19Z)
• CVeDRL: An Efficient Code Verifier via Difficulty-aware Reinforcement Learning [57.24524263804788]
  Code verifiers play a critical role in post-verification for LLM-based code generation.<n>Existing supervised fine-tuning methods suffer from data scarcity, high failure rates, and poor inference efficiency.<n>We show that naive RL with only functionality rewards fails to generate effective unit tests for difficult branches and samples.
  arXiv  Detail & Related papers  (2026-01-30T10:33:29Z)
• Benchmark^2: Systematic Evaluation of LLM Benchmarks [66.2731798872668]
  We propose Benchmark2, a comprehensive framework comprising three complementary metrics.<n>We conduct experiments across 15 benchmarks spanning mathematics, reasoning, and knowledge domains.<n>Our analysis reveals significant quality variations among existing benchmarks and demonstrates that selective benchmark construction can achieve comparable evaluation performance.
  arXiv  Detail & Related papers  (2026-01-07T14:59:03Z)
• Local Coherence or Global Validity? Investigating RLVR Traces in Math Domains [13.626335241662977]
  Reinforcement Learning with Verifiable Rewards (RLVR)-based post-training of Large Language Models (LLMs) has been shown to improve accuracy on reasoning tasks.<n>We investigate the effect of RL post-training on intermediate tokens which are not directly incentivized.
  arXiv  Detail & Related papers  (2025-10-20T23:58:31Z)
• Dissecting Long Reasoning Models: An Empirical Study [94.31064312707211]
  We systematically analyze the roles of positive and negative samples in reinforcement learning (RL)<n>We identify substantial data inefficiency in group relative policy optimization, where over half of the samples yield zero advantage.<n>We investigate unstable performance across various reasoning models and benchmarks, attributing instability to uncertain problems with ambiguous outcomes.
  arXiv  Detail & Related papers  (2025-06-05T11:47:10Z)
• VerifyBench: Benchmarking Reference-based Reward Systems for Large Language Models [53.128374915958624]
  OpenAI o1 and DeepSeek-R1 have achieved remarkable performance in the domain of reasoning.<n>A key component of their training is the incorporation of verifiable rewards within reinforcement learning.<n>Existing reward benchmarks do not evaluate reference-based reward systems.
  arXiv  Detail & Related papers  (2025-05-21T17:54:43Z)
• A Sober Look at Progress in Language Model Reasoning: Pitfalls and Paths to Reproducibility [47.56466996118911]
  Reasoning has emerged as the next major frontier for language models (LMs)<n>We conduct a comprehensive empirical study and find that current mathematical reasoning benchmarks are highly sensitive to subtle implementation choices.<n>We propose a standardized evaluation framework with clearly defined best practices and reporting standards.
  arXiv  Detail & Related papers  (2025-04-09T17:58:17Z)
• The Emperor's New Clothes in Benchmarking? A Rigorous Examination of Mitigation Strategies for LLM Benchmark Data Contamination [18.05548914181797]
  Benchmark Data Contamination (BDC)-the inclusion of benchmark testing samples in the training set-has raised increasing concerns in Large Language Model (LLM) evaluation.<n>To address this, researchers have proposed various mitigation strategies to update existing benchmarks, including modifying original questions or generating new ones based on them.<n>Previous assessment methods, such as accuracy drop and accuracy matching, focus solely on aggregate accuracy, often leading to incomplete or misleading conclusions.
  arXiv  Detail & Related papers  (2025-03-20T17:55:04Z)
• Don't Make Your LLM an Evaluation Benchmark Cheater [142.24553056600627]
  Large language models(LLMs) have greatly advanced the frontiers of artificial intelligence, attaining remarkable improvement in model capacity. To assess the model performance, a typical approach is to construct evaluation benchmarks for measuring the ability level of LLMs. We discuss the potential risk and impact of inappropriately using evaluation benchmarks and misleadingly interpreting the evaluation results.
  arXiv  Detail & Related papers  (2023-11-03T14:59:54Z)
• Deep Reinforcement Learning at the Edge of the Statistical Precipice [31.178451465925555]
  We argue that reliable evaluation in the few run deep RL regime cannot ignore the uncertainty in results without running the risk of slowing down progress in the field. We advocate for reporting interval estimates of aggregate performance and propose performance profiles to account for the variability in results.
  arXiv  Detail & Related papers  (2021-08-30T14:23:48Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.