RL Post-Training

RL: Reinforcement Learning

Follow this topic Ask about this topic

Momentum

49 papers in the last four weeks, up 206% on the four weeks before. 0.5% of all new papers.

Jul 13Week of Sep 28

Latest papers 309

Oct 8, 2026cs.CL

MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement

Reinforcement learning (RL) is the central training paradigm for advancing large foundation models towards self-improvement. This report introduces the MiMo-V2.6 series, an omni-modal family that pushes the frontier of model intelligence by scaling RL compute. Prior to RL, we conduct mid-training on a broad multimodal corpus to provide ample exploration space, and build a solid infrastructure on the pretrained hybrid-SWA architecture to support subsequent scale-up. We scale RL compute along three dimensions: (1) larger batches and higher throughput, with an asynchronous training that consumes 1,568 samples and 2.7-3.7B tokens per step at context lengths of up to 1M; (2) more diverse and complex environments, spanning code, general, visual, and cyber domains under a mixture of agent harnesses; and (3) more grader compute, via groupwise agentic grading that yields more accurate reward signals for long-horizon tasks and steers the model towards shorter, more token-efficient solutions. To keep training stable at scale, we freeze the MoE router and establish a multi-layer defense against reward hacking. We further build infrastructure for mixed-task agentic RL, including a unified trajectory representation, high-concurrency multi-framework rollout, decoupled control and data planes, and training-inference consistency. We open-source the training dynamics, RL environments, and RL framework to facilitate reproduction and further research on scaled RL and model self-improvement.

Xiaomi LLM-Core Team, :, Zongming Qiao +147

Reinforcement Learning Language Model Self-Improvement RL for Language Models RL Post-Training Agentic RL

Oct 8, 2026cs.LG

GRPODropout: Less is More for Online Reinforcement Learning Rollouts

Reinforcement learning (RL) methods such as GRPO substantially improve large language model reasoning but often suffer from policy entropy collapse: the loss of sampling diversity weakens exploration and limits further improvement. Existing methods address this issue either through algorithm-level interventions, such as reward modification and entropy/KL regularization, or through token-level reweighting. We investigate a complementary perspective: entropy collapse can also be mitigated by changing which generated rollouts contribute to policy updates. Under the same sampling budget, not all rollouts contribute positively to an update, and selectively excluding some can improve learning. To address this, we propose GRPODropout: before the standard update, we use a simple strategy that selectively removes a small number of high-probability positive-advantage rollouts and recenters the retained advantages. To motivate this design, we develop a rollout-level theoretical analysis that guides method design and threshold selection. The method changes only rollout usage, and adds negligible computational overhead. Experiments show higher accuracy than original GRPO and higher actor entropy while using fewer rollout samples for updates, illustrating "less is more." This work provides insight into RL rollout usage: removing some rollouts can improve performance. Code is available at https://github.com/hexuandeng/GRPODropout/.

Hexuan Deng, Zihao Yan, Xuebo Liu +8

Institute of Computing and Intelligence, Harbin Institute of Technology (Shenzhen) · Beijing Zhongguancun Academy · XinzhuAI

Reinforcement Learning RL for Language Model Reasoning RL Post-Training Group Relative Policy Optimization

Oct 8, 2026cs.CL

Residual Advantage: Student-Relative Teacher Guidance for RL with Verifiable Rewards

Reinforcement learning with verifiable rewards (RLVR) and on-policy distillation (OPD) have become two main paradigms for post-training reasoning models. RLVR gives each response a single outcome label, leaving the steps inside it without separate credit. OPD provides token-level guidance at student-visited prefixes, but its pointwise signal does not directly reflect the pattern of teacher--student disagreement across the vocabulary. Dense, unbounded log-ratio supervision can amplify the teacher's influence, yet a strong solver is not necessarily a suitable guide when the student's solution paths depart from the teacher's. We propose Residual Advantage (\RA{}), which treats the teacher--student probability residual as a bounded one-step reward, subtracts the corresponding state value under the student policy to form a standard advantage, and centers the result within each response before adding it to the verifier advantage. The guidance term has zero mean within each response, so the verifier advantage remains the response's mean label and the teacher only redistributes credit among the steps within it. \CoRA{} further updates a teacher LoRA with verifier advantages on the same scored student batch and uses the updated teacher in the next iteration's residual, adapting guidance to the student's attempts. With Qwen3-1.7B-Base and Qwen3-4B-Base students and a Qwen3-8B teacher, \RA{} combined with GRPO or REINFORCE++ improves the underlying sequence-advantage algorithm in all 24 comparisons on three mathematical benchmarks, raising macro Avg@8 by 1.7--3.6 points and Pass@8 by 3.9--6.3 points. Both combinations surpass teacher-only OPD, and \CoRA{} adds a further 1.0--1.5 Avg@8 points.

Xiaobing Chen, Zhiqi Pang

Harbin Engineering University · Tencent · Harbin Institute of Technology

RL for Language Model Reasoning RL Post-Training Credit Assignment in RL Reinforcement Learning with Verifiable Rewards On-Policy Distillation

Oct 8, 2026cs.LG

How to post-train on a surrogate: Envelope sampling mitigates reward hacking

Large language models (LLMs) are commonly post-trained against LLM judges and other cheap surrogates because the true reward, such as human preference, is too expensive to query at scale. This practice often leads to reward hacking, where reinforcement learning against a miscalibrated surrogate leads to undesirable side effects. In this work, we study a setting in which a small number

n

of model outputs are annotated with ground-truth labels (e.g., from expert review) and used to recalibrate the LLM judge before optimizing against it. Prior approaches to judge recalibration are costly or heuristic, and it is known that on-policy sampling fails when the surrogate is miscalibrated on a rare set of outputs. In this work, we propose envelope sampling, a theoretically-grounded method for judge recalibration that seeks to minimize an upper bound on the regret of the post-trained model under the assumption that the human reward and re-calibrated reward lie in an

L^2

ball around the judge. We give practical algorithms to sample from the envelope by rejection or by fine-tuning against a modified reward, and experiments on clinical note generation and on a controlled sycophancy task show that recalibrating on envelope samples mitigates reward hacking where recalibrating on base-model samples does not.

Sanjit Dandapanthula, Shuvom Sadhuka, Samir Khan +3

Carnegie Mellon University Department of Statistics and Machine Learning Department · Abridge AI · MIT EECS +2

RL Post-Training Reward Hacking LLM-as-a-Judge Language Model Calibration

Oct 8, 2026cs.LG

Measuring and Mitigating Solution Mode Collapse in RLVR

A language model (LM) can usually answer the same question in more than one way, but reinforcement learning with verifiable rewards (RLVR) is indifferent to which correct answer a model produces. A solution will earn the same reward whether it is the thousandth copy of a familiar answer or one the model has never produced before. Yet, there is potential value in having the model retain multiple correct solutions as it is trained. For instance, multiple modes may give users a choice and provide problem-solving strategies that improve overall model performance. Here, we introduce ModeBench, a benchmark of multi-solution tasks in which the verifier returns both correctness and mode discovered. We then use ModeBench to measure how solution diversity changes under RLVR post-training. We find that RLVR post-training concentrates probability onto fewer correct modes even as accuracy holds or improves, and moreover, that frontier models are already highly concentrated. We then introduce our solution, Re:Max, which stores one verified example per discovered mode in a replay buffer and trains on those stored modes uniformly. A solution found once is, therefore, practiced as often as one found repeatedly. Across three model scales, two RL objectives, and harder task constructions, replay improves both how often a policy succeeds and how many different ways it can succeed.

Liv G. d'Aliberti, Marwa Abdulhai, Sofiia Druchyna +2

Princeton University

Mode Collapse RL Post-Training Reinforcement Learning with Verifiable Rewards Experience Replay

Oct 7, 2026cs.LG

CERO: Where and When to Allocate Rollouts for RL Post-Training

Adaptive rollout methods for group-relative reinforcement learning typically allocate a fixed per-update budget across prompts. We instead study how to coordinate a finite rollout budget over the entire training horizon. We formulate this problem using a concave surrogate utility of cumulative prompt exposure and introduce CERO, an online primal dual scheduler for prompt admission and budget pacing. In our experiments, each admitted prompt receives a fixed-size response group. CERO instead adapts which prompts are selected, how often they are revisited across rounds, and how many groups are generated in each round. A compact Fenchel representation linearizes the dependence on cumulative exposure, while projected online gradient descent updates prompt-specific supporting slopes and a shared budget price using reward-variation feedback and budget deviations. We establish pathwise guarantees for the surrogate allocation objective against fixed-rate and same-path time-varying benchmarks, with explicit terms for proxy discrepancy and rate variation. Under matched training-response budgets, CERO attains the highest avg@16 macro-average on each of three backbones across five mathematical reasoning benchmarks. Mechanistic analyses link CERO's prompt choices to within-group reward contrast, while multi-seed ablations show gains from adaptive pacing over both uniform and preset spending schedules.

Yiming Zong, Yige Wang, Xing Hu +2

Department of Industrial Engineering & Decision Analytics, Hong Kong University of Science and Technology · Innovation and Information Management, The University of Hong Kong · Faculty of Engineering & Faculty of Business and Economics, The University of Hong Kong

Reinforcement Learning RL Post-Training RL for Language Model Reasoning Group Relative Policy Optimization Primal-Dual Optimization

Oct 7, 2026cs.LG

COPC: Coupled Off-Policy Correction for Asynchronous LLM Reinforcement Learning

Asynchronous RL accelerates large language model post-training by decoupling rollout generation from optimization, but trains on stale trajectories. Existing methods primarily correct token-level policy mismatch through importance-ratio control in the actor objective. We show that this \emph{policy-side correction} alone is insufficient: advantage estimates also inherit mismatch from behavior-policy continuations, which we term \emph{advantage staleness}. We derive exact bias and variance decompositions for a general two-channel actor update, revealing nonseparable coupling between policy-weight and advantage-estimation errors: their interaction induces multiplicative bias terms, while squared policy weights amplify advantage uncertainty in gradient variance. This motivates the hypothesis that policy- and advantage-side correction should be coordinated. We introduce Coupled Off-Policy Correction (COPC), an actor--critic method combining token-level ratio masking with two-sided clipped-ratio weighting of TD residuals for return and advantage estimation. Joint parameter sweeps across staleness levels support this hypothesis: the effect of one correction parameter depends on, and can reverse with, the other. COPC achieves the highest reported performance on tool-integrated mathematical reasoning and search, outperforming the strongest reported asynchronous baseline in each setting. It also offers a broad high-performing parameter region and improved training stability. In search, COPC remains stable throughout training, while most evaluated asynchronous baselines collapse late in training. These gains persist at 64-step policy staleness. COPC adds minimal step-time overhead over asynchronous PPO and retains a

1.7\times

step-time speedup over synchronous PPO.

Zicheng Hu, Zhijian Zhou, Xuan Zhang +7

Fudan University · Meituan · East China Normal University

Asynchronous RL RL Post-Training Off-Policy RL RL for Language Model Reasoning

Oct 6, 2026cs.LG

Reinforcement Learning for Hierarchical Reasoning Rewards: Minimax-Optimal Rates with Transformers

Reinforcement learning (RL) has become a standard tool for post-training language models on reasoning tasks, where the policy is updated by reward feedback while exploring the space of responses. Despite its empirical success, theoretical understanding of RL post-training remains limited, in particular of why on-policy exploration combined with a neural reward model is effective. In this paper, we address this question by modeling the reward as a hierarchical function on the response space: the reward consists of infinitely many local components, each of which becomes relevant only after the preceding ones have been resolved. We show that a natural Transformer-based actor--critic algorithm, which alternates between sampling from the current KL-regularized policy, fitting a Transformer critic to the observed rewards, and updating the policy, achieves the minimax optimal rates in the query budget and in the regularization strength up to logarithmic factors, and is minimax optimal for a fixed number of prompts. In contrast, we prove that sampling from the fixed reference distribution, as in offline reward modeling, can limit regret decay to a logarithmic rate. These results show that on-policy exploration progressively zooms in on the region where the reward is concentrated, and quantify its benefit for RL post-training.

Naoki Nishikawa, Taiji Suzuki

The University of Tokyo, RIKEN AIP

Reward Modeling Transformer-Based RL RL for Language Model Reasoning RL Post-Training

Oct 6, 2026cs.LG

TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models

Reinforcement learning (RL) for post-training large language models (LLMs) incurs substantial computation and memory overhead during rollout generation, which motivates low-precision rollout for efficient RL training. However, existing FP4 RL methods suffer from a key limitation: they primarily optimize quantization accuracy on the training and rollout paths independently rather than directly reducing the discrepancy between the two quantized execution paths. In this work, we propose TRACE (Train-Rollout Quantization Alignment via Compact GuidancE), an FP4 quantization framework for RL training of Mixture-of-Experts (MoE) language models that addresses the limitation of existing FP4 RL methods. TRACE incorporates rollout-guided quantization-aware training that uses rollout-side quantization outcomes to guide training-side FP4 rounding decisions, directly reducing train-rollout discrepancy. Moreover, TRACE adopts an efficient quantization-information caching scheme that selectively retains mantissa and scale information from deeper layers to reduce the storage and communication overhead introduced by rollout guidance. We evaluate TRACE on four large-scale MoE language models across reasoning, coding, and long-horizon RL tasks. Our results demonstrate that TRACE enables joint FP4 weight/activation and FP4 KV-cache rollout with RL performance comparable to BF16 rollout, while achieving up to 5.4xrollout speedup and strong final FP4 performance compared with post-hoc FP4 quantization of BF16-trained policies.

Xin Wang, Hao Yu, Zhengyang Zhuge +9

Alibaba Token Hub, Alibaba Group · Ohio State University

LLM Quantization Quantization-Aware Training LLM Inference Acceleration Mixture-of-Experts Quantization RL Post-Training

Oct 5, 2026cs.LG

What pass@k Cannot Measure: Evaluating Diversity and Capability Retention after Post-Training

pass@

k

, the fraction of problems a model solves within

k

sampled attempts, is the field's default protocol for deciding whether reinforcement-learning (RL) post-training on verifiable rewards improved a model. At the population level, pass@

k

depends only on a problem's probability of a correct sample, with no term for how it is distributed across outputs. We show this gap is not academic. Training Qwen2.5-1.5B-Instruct on grade-school math with Group Relative Policy Optimization (GRPO) and with rejection-sampling fine-tuning (RFT, training on the model's own shortest verifier-passed rollout) moves three complementary diversity measures (token-level entropy, answer-level entropy, unique answers per prompt) in opposite directions, with zero overlap across three seeds per arm. The gap survives restricting to verifier-correct completions only (lexical diversity among correct solutions is 15% lower for GRPO, after controlling for length) and a count-controlled check isolating diversity among incorrect answers alone, ruling out that GRPO's higher accuracy alone explains it. Yet pass@8 and pass@32 show no consistent winner on GSM8K, and a hard MATH-500 subset shows the same pattern: separation only at low

k

. Compared against the starting checkpoint, no trained arm significantly improves hard-problem coverage: RFT is significantly worse, while GRPO is statistically indistinguishable from it - so GRPO's pass@1 edge over RFT reflects a smaller loss relative to Base, not a capability gain, a missing-control issue, not a failure of pass@

k

. On GSM8K, only pass@1, with no role in detecting diversity by construction, separates the arms cleanly, rewarding the arm whose correct solutions are least diverse. We argue this is a concrete instance of a standard evaluation protocol missing a property it is routinely used to certify.

Subham Rath, Raj Dandekar, Rajat Dandekar +1

Accenture Strategy & Consulting · Vizuara AI Labs (vizuara.com) · Vizuara AI Labs (vizuara.ai)

LLM Evaluation Language Model Generation Evaluation Reinforcement Learning with Verifiable Rewards RL Post-Training

Oct 5, 2026cs.LG

Structuring MoE Expert Selection for Agentic Reinforcement Learning

Long-horizon LLM agents are frequently implemented using sparse mixture-of-experts (MoE) models, yet the co-design of agentic behavior and MoE structures remains underexplored. In this work, we comprehensively study the connections between agentic post-training and MoE expert selection. In off-the-shelf MoE models, we observe expert selection exhibits a specialized structure that naturally aligns with agentic trajectories. Specifically, expert routing overlaps more between turns where the agent performs semantically similar operations (e.g., READ, UPDATE) than between turns with differing operations. However, standard RL algorithms ignore this specialization, allowing the MoE routing to go uncontrolled during training, which empirically limit task performance and inference efficiency. To address this, we introduce a hierarchical routing control framework for agentic tasks. We explicitly encourage turn-level expert selections to align with agentic operations while regularizing token-level expert selections to maintain local consistency. To resolve stability issues that arise during post-training with the proposed methods, we further introduce an entropy-gated control mechanism. Overall, our routing control framework achieves over 10-point improvements in success rate on all evaluated benchmarks. These results demonstrate that agentic trajectory structure provides an effective signal for optimizing MoE capacity during RL post-training.

Bolian Li, Ting-Yao Hu, Cheng-Yu Hsieh +3

Apple · Purdue University

Agentic RL Expert Routing RL Post-Training Mixture-of-Experts Models

Oct 5, 2026cs.CL

LoGRA: Scaling LLM Reinforcement Learning with Low-Rank Gradient Sketches

Reinforcement learning has greatly advanced the capabilities of large language models, but its memory demands remain a barrier to broader adoption. We introduce LoGRA, an approach to RL post-training that reduces memory by retaining useful learning signals in low-rank gradient sketches. These compact representations support both model updates and efficient policy synchronization. To prevent overly large updates from disrupting learning, we complement gradient compression with predicted-KL step control, which estimates policy changes before applying each update and adjusts its magnitude accordingly. With all techniques combined, LoGRA reduces average training memory usage by up to 45.7% across reasoning tasks without compromising performance. It also enables stable training of a 27B-parameter model for over 1,100 steps on a single eight-GPU node, where dense Adam runs out of memory, making previously memory-infeasible RL training practical. Code is available in the Molt library.

Shaokun Zhang, Yifan Zhang, Jian Hu +5

NVIDIA

Memory-Efficient Fine-Tuning RL for Language Model Reasoning RL Post-Training

Oct 5, 2026cs.RO

How (and How Not) to Use Data Augmentation in VLA Post-Training

Vision-language-action (VLA) models currently demonstrate strong performance in a wide range of real-world robotics tasks. However, they often still lack the generalization ability to handle large visual out-of-distribution shifts. Post-training of VLAs with reinforcement learning (RL) has been shown to benefit robustness, but significant room for improvement remains. In this work, we systematically study the effect of image augmentation on VLA post-training. We find that it is crucial to augment only the critic module during RL updates, while leaving the actor's input clean during both rollouts and updates. For

π_{0.5}

and GR00T N1.5 this raises out-of-distribution success on LIBERO-Plus by

7.8

and

10.0

points respectively, while augmenting the actor collapses training entirely. We investigate a range of augmentation types and strengths, and provide practical recommendations for improving generalization in VLA post-training.

Bram Grooten, Joaquin Vanschoren

TU Eindhoven

RL Post-Training Robustness of VLA Models

Oct 5, 2026cs.LG

ThunderSyncRL: Lossless Acceleration of Agentic Reinforcement Learning

Language models are moving beyond generating answers to pursuing long-horizon goals in interactive environments. Post-training these agents requires long, heterogeneous trajectories, and synchronous systems leave learner engines idle until rollout and verification finish. To squeeze out these pipeline bubbles, asynchronous training overlaps rollout and learning across updates, but comes at the cost of policy staleness. We introduce ThunderSyncRL, which starts gradient computation as soon as all required inputs are fixed, without policy staleness. For group relative policy optimization (GRPO), ThunderSyncRL computes each trajectory's score gradient as soon as the reward for that trajectory arrives, without waiting for the group. For on-policy distillation (OPD), it computes gradients for each completed agentic turn's teacher-scored actions while tool calls run in the sandbox. We prove that gradient streaming produces the same GRPO and OPD updates as batch-synchronous training, without changing either objective. On SWE-bench Verified and Terminal Bench 4.0, we train models to the same performance up to

1.9 \times

faster than synchronous training. With zero policy staleness, ThunderSyncRL also outperforms asynchronous training at a fixed budget by up to

2.47

percentage points.

Seil Kang, Hangoo Kang, Tarun Suresh +4

Stanford University · Yonsei University · Korea University +1

Agentic RL On-Policy Distillation RL Post-Training

Oct 4, 2026cs.LG

Erased, Rerouted, or Rescaled? Post-Training and the Causal Quotient of a Language Model's Belief State

What happens to information a pretrained model already encodes when post-training no longer rewards using it? The common language of representation compression conflates three fates: information may be erased, rerouted away from the decision while still represented, or rescaled to occupy less variance while still represented and used. We make these fates identifiable in models whose pretraining recovers Bayesian belief states. A reward that reads only a coarse function of the hidden state defines an exact reward-null kernel. The kernel lets us separately measure whether the information remains recoverable, whether decisions causally depend on it, and how much activation variance it occupies. Theory says what is protected: KL-anchored reinforcement learning preserves the reference policy's log-odds among equally rewarded outputs, supervised and unanchored objectives carry no such constraint, and spectral compression implies neither erasure nor loss of use. In controlled worlds, post-training mostly reroutes or rescales reward-null information and leaves it decodable. Without an anchor decisions can stop using it although the representation survives, and with one they keep using it. Erasure appears only under prolonged weight decay, for distinctions that neither reward nor next-token prediction can see. Open language models show the same dissociation: in-context belief geometry stays decodable under late-layer spectral compression, and within-class behavior depends on the anchor. Post-training thus selects a causal quotient of the pretrained belief state: the reward defines decision-equivalence, the anchor and the state update protect part of what it ignores, and optimization decides whether the rest is erased, rerouted, or rescaled.

Weihan Li, Tianshi Zheng, Junhao Wu +1

The University of Tokyo · The Hong Kong University of Science and Technology · RWTH Aachen University +1

Belief Updating in LLMs RL Post-Training Representation Geometry in Language Models

Oct 4, 2026cs.RO

$R^2$ -WAM: Repair-and-Reject Post-Training for World Action Models

World Action Models (WAMs) emerge as a promising foundation for policy refinement by predicting the consequences of sampled actions. However, visually plausible predictions can mislead policy refinement if they fail to reflect the input actions. To address this mismatch, we introduce

R^2

-WAM, a two-stage repair-and-reject post-training framework that first improves the consistency of predicted futures with input actions, then uses these futures to select inferior action samples for negative fine-tuning. The repair stage grounds imagination in observed robot behavior through a kinematic alignment score that measures agreement between predicted and demonstrated motion, enabling the predicted video to faithfully reflect its input actions. Using the repaired video model, the rejection stage compares imagined outcomes of sampled and demonstrated actions, selectively applying negative fine-tuning to samples whose predicted task progress falls below the demonstrated reference by a prescribed margin. Together, the two stages extend video prediction from representation learning to consequence-based policy refinement without additional environment interaction or changes to the inference procedure.

R^2

-WAM achieves 93.8% average success on RoboTwin 2.0 across clean and randomized settings. On the long-horizon real-world Fold Shirt task, it achieves 87.5% average success, compared with 0% for Fast-WAM. Our project page is available at https://r2-wam.github.io/.

Ruiyan Xu, Haisheng Su, Sixu Lin +4

The Chinese University of Hong Kong, Shenzhen · Manifold AI · Southeast University +1

World Action Models Action-Conditioned World Models Model-Based RL RL Post-Training

Oct 1, 2026cs.LG

Asynchronous LLM Post-Training: Group-Mass Capping and Convergence Analysis

Asynchronous reinforcement learning (RL) improves the efficiency of large language model post-training but introduces stale rollouts generated by earlier policies. Theoretical understanding of how this staleness affects convergence and how to mitigate its impact remains limited. We derive a convergence bound for GRPO-style algorithms that explicitly characterizes the tradeoff between the gradient estimator's second moment and bias. For trajectory-level importance-weighted estimators, our analysis shows that once the second moment is uniformly controlled, delay enters the bound through the bias introduced by clipping or rescaling. Guided by this insight, we propose a novel group mass capping GRPO (GMC-GRPO) method, which minimizes a ratio-based bias bound within a class of weighted estimators sharing a common second-moment guarantee. We establish convergence guarantees for asynchronous GMC-GRPO and show that, compared with TIC-GRPO, it improves the threshold dependence of the fourth-order delay term from

O(ε^{-4})

to

O(ε^{-2})

as

ε\to0

, where

1+ε

is the ratio threshold. Under local policy overlap, the delay-dependent term decreases as

G^{-2/5}

after tuning the step size, where

G

is the group size. For fixed behavior and current policies, the bias introduced by group rescaling also vanishes as

G\to\infty

, whereas the bias from trajectory-wise clipping can persist. Experiments across Qwen3 models and reasoning benchmarks demonstrate improved robustness to stale rollouts, with GMC-GRPO achieving the best performance among stable baselines under large rollout delays.

Qijia He, Ruinan Jin, Jun Luo +2

The Ohio State University · Arizona State University

Asynchronous RL RL for Language Model Reasoning Group Relative Policy Optimization RL Post-Training Off-Policy RL

Oct 1, 2026cs.LG

Range-GRPO: Policy Optimization via Pairwise Relations among Reward Intervals

As the use of large language models (LLMs) expands, post-training has become increasingly important for adapting them to downstream tasks. However, obtaining reliable supervision remains costly, especially in domains without reference answers or executable verifiers. LLM-as-a-Judge provides scalable pseudo-rewards for unlabeled responses, but a single point score does not explicitly represent reward uncertainty. This motivates representing pseudo-rewards as conformally calibrated reward ranges. We propose Range-GRPO, a semi-supervised post-training framework that combines limited labeled data with unlabeled prompts. In Group Relative Policy Optimization (GRPO), learning signals depend on relative reward comparisons within each rollout group. The proposed objective compares reward ranges pairwise rather than reducing them to point rewards, allowing interval uncertainty to affect both the magnitude and direction of these signals. Our theoretical analysis characterizes this distinction and shows that the proposed objective recovers the Dr.GRPO advantage when all reward ranges collapse to points. Empirically, Range-GRPO achieves the highest in-distribution and out-of-distribution average performance among the evaluated semi-supervised methods while requiring fewer training resources.

Ryunyi Lee, Kangjun Noh, Somin Kim +2

Yonsei University

Reward Modeling LLM-as-a-Judge Group Relative Policy Optimization RL Post-Training Semi-Supervised Learning

Oct 1, 2026cs.AI

Sharpening Tax in Post-Training

An emerging hypothesis about reinforcement learning (RL) post-training of large language models (LLMs) is that it merely sharpens existing behaviors of a base model, improving single-shot accuracy at the cost of solution coverage. Although this trade-off has been observed in math and coding tasks, it need not extend to agentic tasks, where multi-turn tool use and interaction may require capabilities newly acquired during post-training. Our surprising finding is that pre-trained LLMs, equipped with a light inference harness, can serve as capable agents. Despite far lower accuracy (pass@1), they often surpass their post-trained counterparts in solution coverage (pass@K) given a sufficient test-time budget. We further analyze the underlying mechanism and show that post-training pushes tasks toward two extremes, always solved or never solved, and thereby improves sampling efficiency and consistency at the cost of solution coverage. To measure this cost, we propose Sharpening Tax, a diagnostic metric that quantifies the loss in test-time scalability after post-training. Across 14 base/post-trained model pairs from four families and three agentic benchmarks (42 cases in total), the tax is prevalent in most settings, can be estimated from a few rollouts, and correlates well with other metrics. Finally, we present posterior-tempered group sampling (PTGS), a simple plug-and-play Bayesian sampler that adapts the sampling temperature per prompt to its estimated difficulty. Applied during RL training in two agentic environments, PTGS pays a smaller tax than the fixed-temperature baseline, solving more tasks under repeated sampling while also improving single-shot accuracy.

Changdae Oh, Qi Zeng, Qi Qi +7

Meta Superintelligence Labs · University of Wisconsin–Madison · Work done at Meta +2

Agentic RL Test-Time Scaling Adaptive Sampling RL Post-Training

Oct 1, 2026cs.AI

Rethinking Probability-Based Reinforcement Learning From Posterior Concentration

Verifier-free reinforcement learning with probability-based rewards offers a promising way to train LLMs on general reasoning tasks where external verifiers are unavailable. Yet the reliability of these rewards, especially in long-horizon reasoning, remains underexplored. This work identifies a length-dependent failure mode of probability rewards, which we call the Posterior Concentration Phenomenon (PCP). We show that the probability of a reference answer conditioned on a reasoning trace often collapses to a low-variance interval as the trace becomes lengthy. This phenomenon results in nearly indistinguishable rewards, which, under GRPO-based settings, makes probability-based policy optimization unstable and inefficient. Motivated by this, we propose Reinforcement Learning with Concentration-aware Posterior Rewards (RLCPR), a verifier-free RL framework to explicitly account for PCP for better optimization stability and token efficiency. It has two components: uncertainty-aware data sampling, which reduces concentration-prone rollouts before generation, and concentration-aware regularization, which penalizes unnecessarily long traces when posterior rewards collapse. Extensive experiments show that, alongside higher token efficiency, RLCPR outperforms the state-of-the-art verifier-free RL baseline by up to 4.0% on six of seven benchmarks, including general-domain and mathematical reasoning challenges.

Shiu-Hong Kao, Yubo Zhao, Zhenyu Tian +3

National University of Singapore · The Hong Kong University of Science and Technology · University of Science and Technology of China

Reinforcement Learning RL for Language Model Reasoning RL Post-Training

Oct 1, 2026cs.LG

Does Scaling Reinforcement Learning Really Require More Training?

Scaling reasoning typically spends more compute on reinforcement learning (RL) or on inference. We show that a completed RL training history can yield policies stronger than the checkpoints visited by its optimizer. We call this policy-space scaling: expanding the deployable policy set accessible from a fixed RL history, without extending training or increasing per-query inference computation. We instantiate it with SURGE (Scaling Up RL Gradient-free via Eigenspace fusion). SURGE combines two checkpoints from the same RL run: a high-accuracy anchor and a competitive donor that generates shorter responses. It expresses both checkpoints as changes from their shared initialization, then spectrally decomposes the anchor's update to retain its dominant component and incorporate the donor's complementary component. With a fixed target for how much of the anchor update to retain, SURGE determines the block size from the weights without testing candidate policies. We evaluate two 1.5B mathematical-reasoning histories, DeepSeek and Nemotron, and one 7B coding history, OLMo. SURGE improves benchmark-average accuracy over both input checkpoints while using fewer reasoning tokens than the anchor. It reaches 54.17% on DeepSeek AIME24 against a measured native maximum of 50.83%, and 83.7% on OLMo HumanEval+ against 82.8%. These gains exceed the observed training curves. Geometric controls support the importance of RL-update structure beyond weight displacement or token reduction alone. Each constructed model runs as a single policy. Our findings identify stored RL history as a reusable scaling resource: the capability available from a training run need not end at its best checkpoint.

Bangji Yang, Jiajun Fan, Hongbo Ma +2

University of Illinois at Urbana-Champaign · Tsinghua University

RL for Language Model Reasoning RL Post-Training Model Merging

Oct 1, 2026cs.LG

Cross-Benchmark Transfer from RL on Agentic Coding Tasks

Coding agents often fail in the last mile: they build most of a feature but drop a requirement, test only the cases their implementation already handles, break behavior that was supposed to stay intact, or validate against an unchecked assumption. We ask whether reinforcement learning (RL) on expert-built agentic coding tasks closes this gap, and whether what the agent learns transfers beyond the training distribution. We post-train Kimi K2.7 Code, a 1T-parameter (32B active) open-weight mixture-of-experts model, with RL alone on 1,700 tasks: 1,000 repository tasks graded by hidden fail-to-pass tests and by pass-to-pass tests of existing behavior, and 700 terminal tasks graded by expert-written hidden verifiers. The reward is the fraction of target checks passed and drops to zero if any pass-to-pass test fails. One epoch of GSPO on a rank-32 LoRA adapter improves pass@1 on each of the six external benchmarks we evaluated, across three agent harnesses: SWE-Bench Pro (60.1 to 64.8), DeepSWE (31.0 to 43.4), Terminal-Bench 2.1 (67.4 to 82.0), Terminal-Bench 3 (1.4 to 12.1), Terminal-Bench 4 (0.0 to 7.6), and SWE-Marathon (5.0 to 25.0). Pooled over the five independent task sets (Terminal-Bench 4 revises Terminal-Bench 3), the improvement is significant (p < 0.001), and it remains significant on the three sets released after the training data was collected (p = 0.004); the model also improves under both harnesses never used in training. Median trajectories on DeepSWE and Terminal-Bench 3 are 24-35% shorter in agent steps. The base model's failed DeepSWE runs are mostly near-misses, and on the tasks the trained model newly solves, paired trajectories show it avoiding each of the four failure modes above.

Sushant Mehta, Logan Ritchie, Edwin Chen

Surge AI

RL for Code Generation AI Agent Benchmarks Agentic Code Generation RL Post-Training

Sep 30, 2026cs.LG

No Task Vector Is an Island: A Comprehensive Study on the Composability of Task Vectors from On-Policy Distillation

Task vectors provide a simple mechanism for composing learned capabilities through model merging. However, the composability of task vectors produced by on-policy distillation (OPD) remains largely unexplored. OPD trains a student using teacher feedback on student-generated trajectories, yielding parameter updates that differ from those produced by the teacher model, usually by reinforcement learning (RL). We therefore ask whether OPD task vectors can complement their RL teacher updates and compose effectively across tasks. Across five domains and two model architectures, we find evidence for both forms of composability. Within a task, merging OPD and RL task vectors can outperform both constituent models, even when the OPD student is weaker than its RL teacher. Across tasks, OPD task-vector compositions achieve higher average scores than corresponding RL compositions in seven of eight backbone-merging-rule comparisons. Parameter-space analyses reveal substantial non-collinearity between OPD and RL updates. Experiment in CODE domain on SMOLLM3-3B shows that the combined direction outperforms either constituent direction at the tested global update norm, supporting directional complementarity in this configuration. Across tasks, OPD updates also show lower overlap among the top-10% feed-forward channels ranked by update energy. Together, these results show that weaker standalone performance does not imply weaker task-vector composability. OPD task vectors can complement stronger RL teacher updates and combine effectively across tasks, highlighting composability as a distinct property for understanding and evaluating post-training updates.

Jingang Zhou, Feiyu Han, Han Zhu +6

Institute of Automation, Chinese Academy of Sciences · University of Chinese Academy of Sciences · Ant Group

Task Vectors Task Vector Merging On-Policy Distillation RL Post-Training Model Merging

Sep 30, 2026cs.LG

Trust the Critic More

Standard language model RL algorithms credit every token of a long rollout with the same advantage determined by the terminal reward. Actor-critic methods can provide finer-grained credit assignment, but learned critics are generally considered too inaccurate to trust when training LLMs with RL. In recent works, even when a critic is present, it is used only for baseline estimation, so every trajectory must be rolled out to its terminal reward. We introduce Actor-Critic with Action Chunking (AC2) that removes the need to roll every trajectory to completion. AC2 instead assigns credit to action chunks: short continuations of prefixes of past trajectories. A learned critic scores the state reached at the end of each action chunk, allowing the policy to update without observing a terminal reward. We make critic-based credit assignment reliable through three design choices. First, we introduce local readiness which uses critic-based updates on a problem only when the critic is sufficiently accurate on that particular problem. Second, when available, we provide the critic with a reference solution from a previous successful rollout. Third, we assign credit over action chunks of 10k tokens rather than individual tokens, giving the critic a more meaningful portion of the trajectory to evaluate. We train Qwen3-4B on FineProofs-RL using AC2 and evaluate on IMO-ProofBench. AC2 exceeds GRPO's peak validation score of 18.5% using 2.5x fewer decoding FLOPs. This gain comes from two sources, (1) AC2 requires 25% fewer training steps to reach this score, and (2) each step generates fewer tokens because the policy does not need to continue every trajectory to completion. Conceptually, we demonstrate that we can remove the need to roll out every trajectory to completion, opening up a large previously unexplored design space for LLM RL algorithms.

Kaiyue Wen, Luke Bailey, Arvind Mahankali +1

Stanford University

RL for Language Model Reasoning Credit Assignment in RL RL Post-Training Action Chunking Actor-Critic Methods

Sep 30, 2026cs.RO

PRICE the Action Chunks: Physical Relational Credit Assignment for Embodied Reinforcement Learning

Outcome-based reinforcement learning (RL) post-trains vision--language--action policies using terminal success signals, but assigns the same trajectory-level advantage to every action chunk. A failed episode can thus penalize useful early actions as if they caused the failure. Existing approaches seek finer-grained feedback through learned evaluators, adding task-specific supervision or additional model training. We explore, for the first time to our knowledge, whether physical relations across trajectories can provide action-chunk credit in embodied RL from terminal outcomes alone, without an auxiliary evaluator. The key insight is that rollouts reaching corresponding physical situations can serve as references for one another: their terminal outcomes provide evidence for assessing local progress. We introduce Physical Relations for Inferring Credit from Episodes(PRICE), with two components: (i) a physical relational graph that pools current and historical outcomes at corresponding chunk boundaries to estimate success potentials; and (ii) confidence-gated credit assignment that uses changes in these potentials to refine trajectory-level supervision. Our analysis connects oracle potential changes to the terminal-success objective and provides a finite-sample directional bound for outcome-independent evidence pools. Independent continuation tests show that PRICE's retained credits align with local progress, while experiments on LIBERO, RoboTwin 2.0, and real robots demonstrate improved task success over outcome-based baselines and faster learning.

Yangang Zou, Jiajun Lu, Weitao Zhou +6

School of Data Science, Fudan University · Shanghai Innovation Institute · Beihang University +4

Credit Assignment in RL Step-Level Credit Assignment in RL Robotic RL Credit Assignment RL Post-Training

Sep 30, 2026cs.LG

Mitigating the Length-Scaling Tax with Online Distillation

Length scaling during reinforcement-learning (RL) post-training is often viewed as a sign of improved reasoning ability, especially on difficult problems, but may also make responses to already-solved problems unnecessarily verbose. We quantify this side effect as the length-scaling tax (LST): excess response length on already-solved queries without a commensurate accuracy gain. To mitigate LST, we propose Length Self-Distillation (LSD), which routes solved prompts to on-policy distillation and retains the original RL objective for unsolved prompts. LSD uses an exponential moving average of the online policy as its teacher, requiring no external model. We find that LSD achieves comparable or better performance than RL across multiple variants, while substantially curbing response-length growth on easy queries. LSD reduces LST from 19.0% to -3.7% on single-turn reasoning and from 31.4% to 13.7% on multi-turn agentic tasks, demonstrating that LSD effectively preserves concise response patterns on easy queries while supporting efficient exploration on difficult queries during RL post-training.

Xu Wan, Wenyue Xu, Shengjie Zhao +1

ByteDance Seed · Tongji University · Peking University

RL for Language Model Reasoning On-Policy Distillation RL Post-Training RL for LLM Agents

Sep 30, 2026cs.LG

Explicit Trajectory Diversity for RL-Based Post-Training of LLM Agents

LLM agents often admit multiple high-quality solutions to the same task, differing in reasoning structure, tool-use pattern, or interaction trajectory. Yet existing notions of diversity in LLM post-training are mostly implicit, arising from general stochasticity and regularization mechanisms rather than explicitly targeting task-relevant behavioral variation. While such implicit diversity can be useful, it does not directly specify which forms of behavioral variation should be encouraged for a given task. In this work, we study explicit trajectory diversity in RL-based post-training for LLMs. Our key idea is to define diversity through user-specified, task-specific trajectory descriptors, which map each sampled trajectory to an interpretable behavioral representation, and then measure diversity as a set-level functional over the resulting descriptor matrix. Building on this formulation, we introduce Trajectory-guided Joint Policy Optimization(TJPO), a single-policy framework that optimizes explicit diversity over sampled trajectory groups, avoiding the need for population-based policy training, and instantiate it within group-based policy optimization through trajectory-level learning signals. This design makes the diversity objective both interpretable and controllable. Experiments on Sokoban and ALFWorld show that TJPO improves task-specific trajectory diversity while maintaining competitive task performance. Descriptor and trajectory analyses show that the learned variation follows the specified behavioral dimensions and includes distinct successful strategies. Extra experiment results suggest that explicitly shaping trajectory diversity can help LLM agents satisfy user requirements and remain effective when task conditions change.

Huaiyu Fu, Heng Cao, Hao Wang +2

Microsoft

LLM Agent Training RL Post-Training RL for LLM Agents

Sep 29, 2026cs.LG

What Pretraining and Midtraining Make Learnable from Rewards?

A reward can identify a correct answer while leaving the computation needed for new inputs undetermined. We study how pretraining and midtraining supply the information and computation that make reward adaptation effective. In sequential state computation and contextual memory, we characterize mechanisms that agree on every training reward yet demand different held-out answers. Task-independent source observations resolve this ambiguity. We construct finite sampled Adam paths from specified random initializations through source prediction and reward adaptation in the same parameters, proving how prediction acquires execution or retrieval and rewards learn their task-specific use. Experiments with pretrained Qwen2.5 checkpoints test this division of labor. Across eight worlds, Sequential models trained with correct source and first-operation supervision reach 82.61% success, versus 44.15% for a private-random source control. Memory replay preserves retrieval during reward adaptation, and an independent eight-world confirmation achieves 75.32% task success versus 49.86% after matched alternative-retrieval training. GSM8K and HotpotQA separate accuracy at reward entry, subsequent gain and final performance. Together, these results connect information acquisition, executable computation and reward-guided task learning.

Chiwun Yang, Xiaoyu Li

City University of Hong Kong · University of New South Wales

RL for Language Model Reasoning Continued Pretraining RL Post-Training

Sep 29, 2026cs.CV

HaPRL: Human-Anchored Process Reinforcement Learning for Visual Search Agent

Multi-turn visual search agents answer questions about high-resolution images by iteratively deciding where to look. Reinforcement learning for these agents rewards only the final answer, leaving the search process unsupervised. Consequently, faulty routes in which the reasoning process is erroneous yet the final result is correct arise frequently, which in turn leads to ineffective training, i.e., scaling along the wrong paths. In this paper, we introduce HaPRL, the first framework to reinforce the search process with human search behavior. We first build an annotation platform and collect 1K+ human-annotated data with fine-grained behavioral signals. During training, a carefully designed judge scores each rollout with task-adaptive weights, anchored on the distilled trace of how a human annotator actually searched the same image. Extensive experiments show that HaPRL consistently outperforms outcome-based RL, and early-stage process supervision yields 6.7x more improvement in subsequent outcome-based scaling. Our results also demonstrate the importance of aligning model behavior with human process annotation signals, which offer new insight into the training of foundation models.

Zhangquan Chen, Yaoxin Niu, Xiang An +5

Tsinghua University · Peng Cheng Laboratory · LMMs-Lab +2

Human-in-the-Loop Annotation Process Supervision Visual Search RL Post-Training RL for VLMs

Sep 29, 2026cs.LG

Unlocking the Critic: Reward-Free Policy Optimization for LLM Post-Training

Recent approaches to reinforcement learning (RL) post-training for large language models increasingly remove the critic to reduce training instability and memory overhead. Even where a critic is trained, it is discarded once training ends, although it has learned to predict outcomes. We revisit this trend and show that a pretrained critic's ability to predict future outcomes can make it a valuable asset for efficient long-horizon reasoning. First, we find that instability in critic-based RL for long chain-of-thought reasoning is largely an optimization artifact: keeping policy updates small and low in variance restores stable convergence. Second, a well-pretrained critic estimates the posterior probability of eventual success from later trajectory states and unfinished prefixes. Its predictions provide outcome-derived, dense, per-prefix learning signals that, during policy optimization, require neither completed rollouts, step-level annotations, nor external reward labels. Building on this insight, we introduce Reward-Free Policy Optimization (RFPO), which repurposes a single calibrated, frozen critic as a rollout-level reward, a value baseline for generalized advantage estimation, and a success forecaster for unfinished prefixes. We further show that binarizing the debiased score stops the policy from exploiting the critic's length bias. Binarized, RFPO matches supervised PPO without a single label in the training loop, while cutting compute and memory overhead. This makes RFPO well suited to long-horizon reasoning tasks, where outcomes arrive late and generation dominates cost: because rollouts can be rewarded before they finish, training no longer has to pay for waiting on every trajectory to complete. Our findings challenge the prevailing critic-free paradigm and establish critic-based, reward-free optimization as a scalable and computationally efficient path for LLM post-training.

Hongyang Li, Xiao Li, Caesar Wu +3

University of Luxembourg · Seafill Open-Source Community · Université Paris-Saclay

Reinforcement Learning Value Function Estimation RL for Language Model Reasoning RL Post-Training Policy Learning