Derives an exact unbiased policy gradient for RL post-training of diffusion LLMs via entropy-guided step selection and one-step denoising rewards, achieving state-of-the-art results on coding and logical reasoning benchmarks.
hub Canonical reference
A Survey of Reinforcement Learning for Large Reasoning Models
Canonical reference. 100% of citing Pith papers cite this work as background.
42
Pith papers citing it
Background
100% of classified citations
abstract
In this paper, we survey recent advances in Reinforcement Learning (RL) for reasoning with Large Language Models (LLMs). RL has achieved remarkable success in advancing the frontier of LLM capabilities, particularly in addressing complex logical tasks such as mathematics and coding. As a result, RL has emerged as a foundational methodology for transforming LLMs into LRMs. With the rapid progress of the field, further scaling of RL for LRMs now faces foundational challenges not only in computational resources but also in algorithm design, training data, and infrastructure. To this end, it is timely to revisit the development of this domain, reassess its trajectory, and explore strategies to enhance the scalability of RL toward Artificial SuperIntelligence (ASI). In particular, we examine research applying RL to LLMs and LRMs for reasoning abilities, especially since the release of DeepSeek-R1, including foundational components, core problems, training resources, and downstream applications, to identify future opportunities and directions for this rapidly evolving area. We hope this review will promote future research on RL for broader reasoning models. Github: https://github.com/TsinghuaC3I/Awesome-RL-for-LRMs
hub tools
citation-role summary
background 8
citation-polarity summary
roles
background 8polarities
background 8representative citing papers
Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
RaPO reduces catastrophic forgetting in visual continual learning by shaping rewards around policy drift and stabilizing advantages with cross-task exponential moving averages during reinforcement fine-tuning of multimodal models.
SCORP delivers 10-28% gains in safety and 2-7% in efficiency metrics on WOMD by using dual-path scene conditioning in diffusion planning plus variance-gated group-relative policy optimization for closed-loop stability.
OPD for LLMs suffers length inflation and repetition collapse; StableOPD uses reference divergence and rollout mixing to prevent it and improve math reasoning performance by 7.2% on average.
This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.
S²GR adds stepwise thinking tokens with contrastive supervision on codebook clusters to balance computational focus and ground reasoning paths in generative recommendation.
AsyncOPD shows asynchronous OPD training reaches 1.6-3.8x higher throughput than synchronous baselines with comparable accuracy by using forward-KL estimators and multi-sample Monte Carlo correction for finite teacher caches.
CRPO extends group relative policy optimization with stage-dependent uncertainty modeling and reports a 10.4 percentage point weighted F1 gain over RL baselines across 8 mental health datasets.
FGRPO decentralizes GRPO fine-tuning via adaptive aggregation based on relative performance gain to achieve robust convergence on non-IID data while preserving privacy.
Traj-Evolve combines non-parametric experience retrieval and multi-agent RL with a leave-one-out unification strategy to outperform baselines on lung cancer prediction from up to five years of multimodal EHRs, including in never-smokers.
ADR generates novel verifiable code tasks via atomic decomposition and recombination, outperforming heuristic baselines in originality, difficulty, and downstream RLVR gains across coding domains.
DASD improves math reasoning in LLMs by adaptively directing self-distillation based on per-token entropy to balance exploration and step accuracy, outperforming prior self-distillation and RLVR baselines on six benchmarks.
DMPO approximates forward KL minimization in on-policy RL by aligning the policy to a group-level reward-proportional target distribution, yielding 9-12% relative gains over GRPO on NP-Bench and smaller gains on math reasoning.
Entropy polarity is a signed token-level quantity derived from a first-order approximation of entropy change that predicts whether RL updates expand or contract policy entropy in LLM fine-tuning, revealing an asymmetry between high- and low-probability tokens.
Super-Linear Advantage Shaping (SLAS) introduces a non-linear geometric policy update for RL post-training of text-to-image models that reshapes the local policy space via advantage-dependent Fisher-Rao weighting to reduce reward hacking and improve performance over GRPO baselines.
A training recipe for tool-integrated reasoning models achieves state-of-the-art open-source results on math benchmarks such as 96.7% and 99.2% on AIME 2025 at 4B and 30B scales by balancing tool-use trajectories and optimizing for pass@k during SFT before stable RLVR.
A 7B/8B model trained with decoupled tri-perspective SFT and QA-verified RL matches GPT-4o and approaches GPT-5 on chart-to-code generation benchmarks.
TGR performs manifold-informed latent foresight search to boost trajectory coverage in long-context reasoning tasks by up to 13 AUC points with minimal overhead.
NPR trains LLMs to reason in parallel via self-distilled RL, delivering up to 24.5% performance gains and 4.6x speedups with 100% genuine parallel execution on reasoning benchmarks.
The paper identifies that importance sampling ratios in outcome-supervised RL misallocate credit by creating unbalanced token updates, and introduces ASPO to correct the asymmetry for positive-advantage tokens.
Post-training on reasoning tasks sparks the emergence of specialized attention heads that enable structured computation, with SFT adding stable heads while GRPO uses dynamic activation and pruning tied to reward signals, and controllable think models relying on compensatory heads instead of specific
PointVG-R is a new MLLM that reaches SOTA on pointing localization by 15.86 mIoU points via a geometric reasoning pipeline, EgoPoint-CoT dataset, SFT, RL, and variance-based reward weighting.
Survey mapping RL techniques onto LLM training and highlighting gaps in value-based, off-policy, and bootstrapping methods.
citing papers explorer
-
Reinforcement Learning for Diffusion LLMs with Entropy-Guided Step Selection and Stepwise Advantages
Derives an exact unbiased policy gradient for RL post-training of diffusion LLMs via entropy-guided step selection and one-step denoising rewards, achieving state-of-the-art results on coding and logical reasoning benchmarks.
-
Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning
RaPO reduces catastrophic forgetting in visual continual learning by shaping rewards around policy drift and stabilizing advantages with cross-task exponential moving averages during reinforcement fine-tuning of multimodal models.
-
SCORP: Scene-Consistent Multi-agent Diffusion Planning with Stable Online Reinforcement Post-Training for Cooperative Driving
SCORP delivers 10-28% gains in safety and 2-7% in efficiency metrics on WOMD by using dual-path scene conditioning in diffusion planning plus variance-gated group-relative policy optimization for closed-loop stability.
-
Demystifying OPD: Length Inflation and Stabilization Strategies for Large Language Models
OPD for LLMs suffers length inflation and repetition collapse; StableOPD uses reference divergence and rollout mixing to prevent it and improve math reasoning performance by 7.2% on average.
-
Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning
This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.
-
S$^2$GR: Stepwise Semantic-Guided Reasoning in Latent Space for Generative Recommendation
S²GR adds stepwise thinking tokens with contrastive supervision on codebook clusters to balance computational focus and ground reasoning paths in generative recommendation.
-
Mental-R1: Aligning LLM Reasoning for Mental Health Assessment
CRPO extends group relative policy optimization with stage-dependent uncertainty modeling and reports a 10.4 percentage point weighted F1 gain over RL baselines across 8 mental health datasets.
-
FGRPO: Federated GRPO with Adaptive Aggregation on Non-IID Data
FGRPO decentralizes GRPO fine-tuning via adaptive aggregation based on relative performance gain to achieve robust convergence on non-IID data while preserving privacy.
-
Traj-Evolve: A Self-Evolving Multi-Agent System for Patient Trajectory Modeling in Lung Cancer Early Detection
Traj-Evolve combines non-parametric experience retrieval and multi-agent RL with a leave-one-out unification strategy to outperform baselines on lung cancer prediction from up to five years of multimodal EHRs, including in never-smokers.
-
Combinatorial Synthesis: Scaling Code RLVR via Atomic Decomposition and Recombination
ADR generates novel verifiable code tasks via atomic decomposition and recombination, outperforming heuristic baselines in originality, difficulty, and downstream RLVR gains across coding domains.
-
Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning
DASD improves math reasoning in LLMs by adaptively directing self-distillation based on per-token entropy to balance exploration and step accuracy, outperforming prior self-distillation and RLVR baselines on six benchmarks.
-
Beyond Mode Collapse: Distribution Matching for Diverse Reasoning
DMPO approximates forward KL minimization in on-policy RL by aligning the policy to a group-level reward-proportional target distribution, yielding 9-12% relative gains over GRPO on NP-Bench and smaller gains on math reasoning.
-
Entropy Polarity in Reinforcement Fine-Tuning: Direction, Asymmetry, and Control
Entropy polarity is a signed token-level quantity derived from a first-order approximation of entropy change that predicts whether RL updates expand or contract policy entropy in LLM fine-tuning, revealing an asymmetry between high- and low-probability tokens.
-
Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shaping
Super-Linear Advantage Shaping (SLAS) introduces a non-linear geometric policy update for RL post-training of text-to-image models that reshapes the local policy space via advantage-dependent Fisher-Rao weighting to reduce reward hacking and improve performance over GRPO baselines.
-
Teaching Thinking Models to Reason with Tools: A Full-Pipeline Recipe for Tool-Integrated Reasoning
A training recipe for tool-integrated reasoning models achieves state-of-the-art open-source results on math benchmarks such as 96.7% and 99.2% on AIME 2025 at 4B and 30B scales by balancing tool-use trajectories and optimizing for pass@k during SFT before stable RLVR.
-
The Geometric Reasoner: Manifold-Informed Latent Foresight Search for Long-Context Reasoning
TGR performs manifold-informed latent foresight search to boost trajectory coverage in long-context reasoning tasks by up to 13 AUC points with minimal overhead.
-
Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning
NPR trains LLMs to reason in parallel via self-distilled RL, delivering up to 24.5% performance gains and 4.6x speedups with 100% genuine parallel execution on reasoning benchmarks.
-
When Importance Sampling Misallocates Credit: Asymmetric Ratios for Outcome-Supervised RL
The paper identifies that importance sampling ratios in outcome-supervised RL misallocate credit by creating unbalanced token updates, and introduces ASPO to correct the asymmetry for positive-advantage tokens.
-
Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Training
Post-training on reasoning tasks sparks the emergence of specialized attention heads that enable structured computation, with SFT adding stable heads while GRPO uses dynamic activation and pruning tied to reward signals, and controllable think models relying on compensatory heads instead of specific
-
PointVG-R: Internalizing Geometric Reasoning in MLLMs for Precise Pointing Localization via Visual Chain of Thought
PointVG-R is a new MLLM that reaches SOTA on pointing localization by 15.86 mIoU points via a geometric reasoning pipeline, EgoPoint-CoT dataset, SFT, RL, and variance-based reward weighting.
-
Modularized Reinforcement Learning on LLMs: From MDP Creation to Exploration and Learning
Survey mapping RL techniques onto LLM training and highlighting gaps in value-based, off-policy, and bootstrapping methods.
-
IAPO: Input Attribution-Aware Policy Optimization for Tool Use in Small Multimodal Agents
IAPO is an RL method that aligns model input attributions with a teacher to improve tool-calling in multimodal SLMs, reporting 3% average VQA accuracy gains on Qwen2.5-VL-3B across six tests.
-
When RL Fails after SFT: Rejuvenating Model Plasticity for Robust SFT-to-RL Handoff
Excessive SFT reduces LLM plasticity for RL; Rejuvenation restores it via base-anchored fusion and targeted neuron resets, yielding better RL performance and OOD generalization.
-
GeoMin: Data-Efficient Semi-Supervised RLVR via Geometric Distribution Modeling
GeoMin uses geometric distribution modeling on labeled data to assess self-reward reliability, enabling better performance in semi-supervised RLVR with only 10% of typical annotations.
-
RLVR without Ineffective Samples: Group Prioritized Off-Policy Optimization for LLM Reasoning
POPO uses recency-based prioritized group replay and decoupled off-policy optimization to avoid zero-variance ineffective samples in RLVR, accelerating LLM reasoning finetuning with fewer rollouts.
-
Meta-Cognitive Memory Policy Optimization for Long-Horizon LLM Agents
MMPO introduces Belief Entropy as a self-supervised signal to provide fine-grained supervision for memory policies in LLM agents, outperforming outcome-based RL on long-horizon tasks up to 1.75M tokens.
-
DVAO: Dynamic Variance-adaptive Advantage Optimization for Multi-reward Reinforcement Learning
DVAO dynamically weights multi-objective advantages by rollout-group reward variance to bound magnitudes, add cross-objective regularization, and outperform static baselines on math and tool-use tasks with Qwen models.
-
Global Context or Local Detail? Adaptive Visual Grounding for Hallucination Mitigation
PND mitigates object hallucination in vision-language models via dual-path contrastive decoding that boosts visual evidence and penalizes linguistic priors, yielding up to 6.5% gains on POPE, MME, and CHAIR benchmarks.
-
StaRPO: Stability-Augmented Reinforcement Policy Optimization
StaRPO improves LLM reasoning by adding autocorrelation function and path efficiency stability metrics to RL policy optimization, yielding higher accuracy and fewer logic errors on reasoning benchmarks.
-
POPI: Personalizing LLMs via Optimized Natural Language Preference Inference
POPI distills user preferences into reusable natural-language summaries via a shared inference model and conditions a generator on them, trained jointly with RL to improve personalization quality while cutting context length by up to 10x on benchmarks.
-
Rethinking Entropy Interventions in RLVR: An Entropy Change Perspective
Derives a token-level entropy change approximation revealing four factors, identifies limitations in prior entropy interventions, and proposes STEER which adaptively reweights tokens to mitigate collapse and improve performance on math and coding benchmarks.
-
Next-Generation Agentic Reinforcement Learning Systems Enable Self-Evolving Agents
Current agentic RL systems lack three key components needed for self-evolving agents at scale, requiring new co-designed architectures such as AReaL2.0 to enable policy updates from deployed workloads.
-
Shattering the Autoregressive Curse: Dynamic Epistemic Entropy Orchestrated Erasable Reinforcement Learning for LLMs
E³RL uses dynamic thresholds on epistemic entropy from autoregressive cross-entropy to enable erasable RL in LLM reasoning, reporting 5.349% and 6.514% gains on AIME for 4B and 8B models over prior SOTA.
-
Watch, Remember, Reason: Human-View Video Understanding with MLLMs
This is a survey that frames video MLLM research via a human-view formulation of perceptual representations, memory states, reasoning traces, and predictions, then reviews methods, datasets, benchmarks, and open problems.
-
Improving Small Language Models for Code Generation with Reinforcement Learning from Verification Feedback
RLVR with combined unit-test and static-analysis rewards improves pass@1 by up to 13pp on MBPP for 0.6B-1B models, while single-reward variants can induce shorter but less correct outputs.
-
Entropy-KL Divergence-based Token Masking: A Novel Approach for Selective Fine-tuning of Large Language Models
EKSFT masks high-entropy or high-KL tokens in low-data SFT to preserve pre-trained distribution and improve downstream RL performance on math reasoning tasks.
-
Agentic Reasoning for Large Language Models
The survey structures agentic reasoning for LLMs into foundational, self-evolving, and collective multi-agent layers while distinguishing in-context orchestration from post-training optimization and reviewing applications across domains.