| A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM | ✓ Link | 96.7 | A*-Thought-V2 (Qwen3.6-27B, tau=90) | 2026-09-07 |
| Efficient Agentic Reasoning Through Self-Regulated Simulative Planning | ✓ Link | 96.2 | SR 2 AM-v1.0-30B | 2026-05-21 |
| Nemotron-Math: Efficient Long-Context Distillation of Mathematical Reasoning from Multi-Mode Supervision | ✓ Link | 94.67 | Qwen3-30B-A3B (fine-tuned on Nemotron-Math, high reasoning, Python TIR, full-data training) | 2025-12-17 |
| Xolver: Multi-Agent Reasoning with Holistic Experience Learning Just Like an Olympiad Team | ✓ Link | 94.4 | Xolver | 2025-06-17 |
| Introducing OpenAI o3 and o4-mini | ✓ Link | 93.4 | OpenAI o4-mini | 2025-04-16 |
| Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training | ✓ Link | 93.3 | Qwen3.5-9B (rho=30%) | 2026-10-09 |
| Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning | | 93.2 | Ring-2.5-1T-Zero (High) | 2026-07-14 |
| Unlocking Lossless Speedups in LLMs via Discrete Diffusion | ✓ Link | 93 | Uno (K2-Horizon-7B-Uno) | 2026-09-03 |
| SkillOS: Learning Skill Curation for Self-Evolving Agents | | 92.2 | SkillOS (Qwen3-8B curator) with Gemini-2.5-Pro executor | 2026-05-07 |
| Loop the Loopies! | | 92.09 | Loopie-20B-A2B (Thinking) | 2026-07-17 |
| Solar Open Technical Report | ✓ Link | 91.7 | Solar Open (102B) | 2026-01-11 |
| Recursive Think-Answer Process for LLMs and VLMs | | 90.9 | Qwen3-235B-A22B-R-TAP (MoE, 235B) | 2026-03-02 |
| INTELLECT-3: Technical Report | ✓ Link | 90.8 | INTELLECT-3 | 2025-12-18 |
| Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models | ✓ Link | 90.4 | Nanbeige4-3B-Thinking-2511 | 2025-12-06 |
| QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management | ✓ Link | 90.0 | QwenLong-L1.5-30B-A3B | 2025-12-15 |
| Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models | ✓ Link | 90.0 | Qwen3-30B-A3B-Thinking with PUMA | 2026-05-17 |
| ACC: Compiling Agent Trajectories for Long-Context Training | | 90.0 | Qwen3-30B-A3B-Thinking + ACC | 2026-05-21 |
| Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models | | 89.7 | Nemotron-Cascade (14B, Thinking) | 2025-12-15 |
| GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models | ✓ Link | 89.4 | GLM-4.5-Air | 2025-08-08 |
| Qwen3-Coder-Next Technical Report | ✓ Link | 89.01 | Qwen3-Coder-Next | 2026-02-28 |
| JudgeRLVR: Judge First, Generate Second for Efficient Reasoning | | 89.0 | JudgeRLVR (Qwen3-30B-A3B) | 2026-01-13 |
| Distribution-Aligned Sequence Distillation for Superior Long-CoT Reasoning | ✓ Link | 88.5 | DASD-4B-Thinking (4B) | 2026-01-14 |
| Cluster, Route, Escalate: Cascaded Framework for Cost-Aware LLM Serving | ✓ Link | 88.4 | Stage 1+2 (λ=0.06, QE on C1) | 2026-06-25 |
| Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling | ✓ Link | 88.1 | Falcon-H1R (7B) | 2026-01-05 |
| Does On-Policy Distillation Really Distill? From Noisy Teacher to Self-Improvement | ✓ Link | 87.81 | OPSA (Qwen3.5-9B) | 2026-08-31 |
| Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT Data | | 87.5 | Mixed SFT + RLVR | 2026-08-24 |
| CHIMERA: Compact Synthetic Data for Generalizable LLM Reasoning | | 86.9 | Qwen3-4B-Thinking-2507 (4B) fine-tuned on Chimera (SFT+RL) | 2026-03-01 |
| ReTerm: Recovery Training for Self-Improving Terminal Agents | ✓ Link | 86.7 | ReTerm-9B | 2026-10-13 |
| Learning to Build the Environment: Self-Evolving Reasoning RL via Verifiable Environment Synthesis | | 86.2 | EvoEnv (Qwen3-4B-Thinking-2507) | 2026-05-14 |
| Qwen3 Technical Report | ✓ Link | 85.7 | Qwen3-235B-A22B (Thinking) | 2025-05-14 |
| Self-Organizing Agent Teams Learn to Reason Together | | 84.7 | Self-Organizing Agent Teams (SAT) | 2026-09-19 |
| Lightning Weave: Improving the Accuracy-Efficiency Frontier of Reasoning Models through Capability Composition | | 84.2 | Lightning Weave (Qwen3.5-4B) | 2026-09-13 |
| MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling | | 83.75 | MiniCPM-SALA (9B) | 2026-02-12 |
| Length-Unbiased Sequence Policy Optimization: Revealing and Controlling Response Length Variation in RLVR | ✓ Link | 83.6 | Qwen3-30B-A3B-Instruct (LUSPO) | 2026-02-05 |
| Introspective Diffusion Language Models | ✓ Link | 83.3 | I-DLM (32B) | 2026-04-13 |
| SCALE: Selective Resource Allocation for Overcoming Performance Bottlenecks in Mathematical Test-time Scaling | ✓ Link | 82.92 | Qwen3-32B (SCALE) | 2025-11-29 |
| On-Policy Distillation with Negative-Policy Rollouts | ✓ Link | 81.7 | Qwen3-8B + NP-OPD (OPD 2, alpha=0.5, thinking mode) | 2026-10-06 |
| ASTRA: Automated Synthesis of agentic Trajectories and Reinforcement Arenas | ✓ Link | 81.4 | ASTRA-32B-Thinking-v1 | 2026-01-29 |
| Hunyuan Dense V1 | ✓ Link | 81.1 | Hunyuan-7B-Instruct | 2025-07-30 |
| Reasoning over mathematical objects: on-policy reward modeling and test time aggregation | | 81.06 | Principia-4B | 2026-03-19 |
| ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search | ✓ Link | 80.62 | ZGCM-1 | 2026-09-11 |
| Olmo 3 | ✓ Link | 80.6 | Olmo 3.1 Think 32B | 2025-12-15 |
| How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize Student-Consistent SFT Data | ✓ Link | 80.42 | TESSY | 2026-03-23 |
| ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models | ✓ Link | 79.9 | ThreadWeaver (Qwen3-8B) | 2025-11-24 |
| DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | ✓ Link | 79.8 | DeepSeek-r1 | 2025-01-22 |
| Post-Trained MoE Can Skip Half Experts via Self-Distillation | ✓ Link | 79.8 | ZEDA (GLM-4.7-Flash) | 2026-05-18 |
| Distilling Long-CoT Reasoning through Collaborative Step-wise Multi-Teacher Decoding | ✓ Link | 79.6 | R1-Qwen (32B) with CoRD (heterogeneous teachers) | 2026-05-04 |
| Figure It Out: Improving the Frontier of Reasoning with Active Visual Thinking | ✓ Link | 79.58 | FigR | 2025-12-30 |
| Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information | ✓ Link | 79.4 | Qwen3-30B-A3B + AntiSD | 2026-05-12 |
| ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning | ✓ Link | 79.3 | ThoughtFold (Qwen3-14B) | 2026-06-02 |
| Bridged Self-Distillation for On-Policy Learning | ✓ Link | 79.2 | Qwen3-4B + BSD (Qwen3-32B teacher) (extra data) | 2026-10-19 |
| CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning | ✓ Link | 79.17 | CopT (Qwen3-8B) | 2026-05-19 |
| Better Supervision Is Nearby: Neighborhood On-Policy Self-Distillation | | 78.7 | N-OPSD (Qwen3-8B) | 2026-09-30 |
| Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning | | 77.5 | OASIS (Qwen3-8B) (extra data) | 2026-09-29 |
| On-Policy Self-Distillation without Any Supervision | ✓ Link | 76.94 | u-OPSD Qwen3-8B (thinking) | 2026-08-09 |
| iGRPO: Self-Feedback-Driven LLM Reasoning | | 76.72 | OpenMath-Nemotron-14B + iGRPO | 2026-02-09 |
| Surprising Success, Repeated Failure: Entropy-Guided Credit Assignment for Exploration in LLM Reasoning | ✓ Link | 76.7 | Olmo-3-7B-Think-DPO + EAPO | 2026-09-27 |
| On-Policy Self-Distillation for Reasoning Compression | ✓ Link | 76.3 | OPSDC (Qwen3-14B, 30K budget) | 2026-03-05 |
| Where the Model Changes Its Mind: Hindsight-Divergence Localization for Efficient Reinforcement Learning with Verifiable Rewards | | 75.97 | HDL (Qwen3-8B) | 2026-09-29 |
| KnowRL: Boosting LLM Reasoning via Reinforcement Learning with Minimal-Sufficient Knowledge Guidance | ✓ Link | 75.52 | KnowRL-Nemotron-1.5B (CBRS) | 2026-04-14 |
| Rubric-based On-policy Distillation | ✓ Link | 75.41 | ROPD (Qwen3-4B) | 2026-05-08 |
| OmniOPD: Logit-Free On-Policy Distillation via Speculative Verification | | 75.0 | OmniOPD (Qwen3-4B student, Qwen3-30B-A3B-Instruct teacher) | 2026-05-31 |
| Knowing When to Quit: Diagnosing and Training LLMs to Abort Futile Reasoning | ✓ Link | 74.6 | CaRL-8B (extra data) | 2026-07-31 |
| | 74.4 | Openai-o1 | |
| Not All Tokens Deserve Equal Credit: Counterfactual Sensitivity Credit Reallocation for Long-CoT Reasoning | | 74.4 | CSCR (Qwen3-4B) | 2026-07-30 |
| Efficient Reasoning with Balanced Thinking | ✓ Link | 73.3 | ReBalance (Qwen3-14B) | 2026-03-12 |
| Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling | | 73.0 | RL-guided (Qwen3-4B-Instruct-Thinking) | 2026-06-02 |
| Scaling Embeddings Outperforms Scaling Experts in Language Models | | 72.19 | LongCat-Flash-Lite (68.5B) | 2026-01-29 |
| Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation | ✓ Link | 71.0 | Lightning OPD | 2026-04-14 |
| CoDiQ: Test-Time Scaling for Controllable Difficult Question Generation | ✓ Link | 70.6 | CoDiQ-L3-4B | 2026-02-02 |
| LongCat-Flash Technical Report | ✓ Link | 70.42 | LongCat-Flash | 2025-09-01 |
| LSRIF: Logic-Structured Reinforcement Learning for Instruction Following | | 70.2 | Distill-Qwen-14B (LsrIF) | 2026-01-10 |
| | 70.0 | Openai-o1-mini | |
| JustRL: Scaling a 1.5B LLM with a Simple RL Recipe | ✓ Link | 69.69 | JustRL-Nemotron 1.5B | 2025-12-18 |
| Kimi K2: Open Agentic Intelligence | ✓ Link | 69.6 | Kimi K2 | 2025-07-28 |
| Weak-to-Strong On-Policy Distillation | ✓ Link | 68.9 | W2S-OPD (single-teacher contrast) | 2026-07-28 |
| Revisiting Complete Reasoning Traces for Post-Training | ✓ Link | 68.89 | Qwen2.5-32B-Instruct + segment SFT (s1K-1.1, Both endpoints) | 2026-09-07 |
| Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation | ✓ Link | 68.85 | Waypoint-OPD (Qwen3-8B) | 2026-10-15 |
| Can We Trust the Teacher? Decoupled Credit Direction-Magnitude for Self-Distillation | | 68.33 | DCSD (Qwen3-4B) | 2026-09-28 |
| Self-Distillation Zero: Self-Revision Turns Binary Rewards into Dense Supervision | ✓ Link | 68.3 | SD-Zero (Qwen3-4B-Instruct) | 2026-04-13 |
| Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment | ✓ Link | 67.5 | Qwen-3-14B (fine-tuned with Rank-Surprisal Ratio selection) | 2026-01-20 |
| Calibrating Teacher--Student Discrepancy for On-Policy Distillation | | 67.1 | Cal-OPD (Qwen3-30B-A3B-Thinking-2507 -> Qwen3-4B) | 2026-09-18 |
| Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation | ✓ Link | 66.92 | Qwen3-8B + OPRD | 2026-09-08 |
| LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning | ✓ Link | 66.67 | LYNX (QwQ-32B) | 2025-12-05 |
| Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability | ✓ Link | 66.0 | Qwen3-14B (Math-CoT SFT) | 2026-04-08 |
| TEMPO: Scaling Test-time Training for Large Reasoning Models | ✓ Link | 65.8 | Qwen3-14B-Base (TEMPO) | 2026-04-21 |
| Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models | ✓ Link | 65.4 | Youtu-LLM 2B | 2025-12-31 |
| Rethinking Sample Polarity in Reinforcement Learning with Verifiable Rewards | | 65.2 | A3PO (Deepseek-R1-Distill-Qwen-7B) | 2025-12-25 |
| SAF-OPD: Stable Advantage Fusion for On-Policy Distillation | | 64.79 | SAF (Qwen3-8B) | 2026-07-31 |
| Filter, Then Reweight: Rethinking Optimization Granularity in On-Policy Distillation | ✓ Link | 64.17 | FiRe-OPD | 2026-06-01 |
| In-Context Reinforcement Learning for Tool Use in Large Language Models | ✓ Link | 64.1 | Qwen3-8B (ICRL) | 2026-03-09 |
| Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning | | 63.5 | ERC-GPPO (DeepSeek-R1-Distill-Qwen-7B) | 2025-12-05 |
| Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning | ✓ Link | 63.3 | NPR (Qwen3-4B-Instruct-2507) | 2025-12-08 |
| Influence-Directed Distillation: Solving the Diversity Bottleneck in Sampled-Token On-Policy Distillation | | 63.3 | Qwen3-8B-Non-Thinking + IDA-OPD | 2026-08-30 |
| The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation | ✓ Link | 63.13 | RIDE (Qwen3-4B -> Just-Qwen3-4B) | 2026-09-29 |
| Learning from Your Own Mistakes: Constructing Learnable Micro-Reflective Trajectories for Self-Distillation | | 62.50 | Cold-start + TAPO (Qwen3-8B-Instruct) | 2026-06-17 |
| SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance | ✓ Link | 62.04 | Qwen3.5-35B w/SAGE | 2026-09-24 |
| Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation | | 61.88 | Road MOPD (Qwen3-4B) | 2026-10-09 |
| Weak-to-Strong Generalization via Direct On-Policy Distillation | ✓ Link | 61.2 | R1-Distill-7B (Direct-OPD) | 2026-07-06 |
| Efficient Reasoning on the Edge | | 61.0 | Qwen2.5-7B-Instruct (OT3, dense) | 2026-03-17 |
| Co-Evolving Policy Distillation | | 60.42 | CoPD | 2026-04-29 |
| MOPD-Router: Rethinking Teacher Routing in Multi-Teacher On-Policy Distillation | ✓ Link | 60.42 | MOPD-Router (ExpertAlign, Qwen3-4B-Non-Thinking student, domain-labeled) | 2026-09-25 |
| How2Everything: Mining the Web for How-To Procedures to Evaluate and Improve LLMs | ✓ Link | 59.06 | Qwen3-8B (Instruct) + RL | 2026-02-09 |
| RISE: Recursive Improvement via Self-Extrapolating Policy Distillation | | 58.5 | RISE (logit,Qwen3-8B,cosineschedule) | 2026-09-04 |
| Bootstrapping Exploration with Group-Level Natural Language Feedback in Reinforcement Learning | ✓ Link | 58.49 | GOLF (Qwen-3-8B) | 2026-03-04 |
| Self-Hinting Language Models Enhance Reinforcement Learning | ✓ Link | 58.1 | SAGE (Qwen3-4B-Instruct) | 2026-02-03 |
| Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math | | 57.5 | Phi-4-Mini-Reasoning | 2025-04-30 |
| DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards | ✓ Link | 56.87 | DelTA (Qwen3-14B-Base) | 2026-05-20 |
| MinT: Managed Infrastructure for Training and Serving Millions of LLMs | ✓ Link | 56.8 | Qwen3-8B (LoRA via MinT, GRPO) | 2026-05-13 |
| Search-o1: Agentic Search-Enhanced Large Reasoning Models | ✓ Link | 56.7 | Search-o1 | 2025-01-09 |
| s1: Simple test-time scaling | ✓ Link | 56.7 | s1-32B | 2025-01-31 |
| ARISE: Agent Reasoning with Intrinsic Skill Evolution in Hierarchical Reinforcement Learning | ✓ Link | 56.4 | ARISE + GRPO (Qwen3-4B-Instruct-2507) | 2026-03-17 |
| From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Space | ✓ Link | 56.15 | DSRL (Qwen3-8B) | 2026-04-15 |
| ERNIE 4.5 Technical Report | ✓ Link | 54.8 | ERNIE-4.5-300B-A47B | 2025-06-30 |
| STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability | ✓ Link | 53.8 | STARE-O1 (DeepSeek-R1-Distill-Qwen-1.5B) | 2026-06-17 |
| Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding | | 53.33 | Nemotron-Labs-Diffusion-14B (Quad. SS) | 2026-07-07 |
| InftyThink+: Effective and Efficient Infinite-Horizon Reasoning via Reinforcement Learning | ✓ Link | 52.29 | InftyThink+ (Qwen3-4B-Base, task-only RL) | 2026-02-06 |
| Trust Region On-Policy Distillation | | 52.08 | TrOPD (Qwen3-SFT-1.7B) | 2026-05-31 |
| Small Generalizable Prompt Predictive Models Can Steer Efficient RL Post-Training of Large Reasoning Models | | 51.8 | DSR-7B with GPS | 2026-02-02 |
| GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts | ✓ Link | 51.67 | GlimpRouter (Qwen3-4B SLM + DeepSeek-R1-Distill-Qwen-32B LLM) | 2026-01-08 |
| TREK: Distill to Explore, Reinforce to Refine | | 51.1 | Qwen3-8B + TREK (DeepSeek-V4 proposals) | 2026-07-06 |
| LFM-1B-Math: Can Small Models Be Concise Reasoners? | | 51.04 | LFM-1.3B-Distill (32K output budget) | 2025-06-24 |
| Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs? | ✓ Link | 51.04 | DeepSeek-R1-Distill-Qwen-7B (SFT on D_ug) | 2026-03-25 |
| Approximation of Log-Partition Function in Policy Mirror Descent Induces Implicit Regularization for LLM Post-Training | ✓ Link | 50.83 | PMD-mean (Qwen3-30B-A3B-Base, τ=0.1) | 2026-02-05 |
| ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning | ✓ Link | 50.8 | ResRL (Qwen3-8B) | 2026-05-01 |
| Trust Region Policy Distillation | | 50.42 | Qwen3-8B-Base (TOP-D) | 2026-07-06 |
| Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs | ✓ Link | 50.0 | Phi-4-Mini reasoning | 2025-02-24 |
| Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation | ✓ Link | 50.0 | OPDLM-MATH-8B-Thinking | 2026-06-04 |
| GRAIL: Gradient-Reweighted Advantages for Reinforcement Learning with Verifiable Rewards | ✓ Link | 47.78 | GRAIL (Qwen3-8B) | 2026-06-03 |
| Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoders | ✓ Link | 46.9 | OPD student (JustRL) | 2026-09-28 |
| Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models | | 46.7 | MELT-1.6B | 2026-05-08 |
| IIB-LPO: Latent Policy Optimization via Iterative Information Bottleneck | | 46.6 | I2B-LPO (Qwen3-14B) | 2026-01-09 |
| ESPO: Early-Stopping Proximal Policy Optimization | | 46.28 | ESPO (DeepSeek-R1-Distill-Qwen-7B) | 2026-05-28 |
| The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping | ✓ Link | 45.78 | Qwen3-8B + MEDS | 2026-04-13 |
| Demystifying On-Policy Distillation: Roles, Pathologies, and Regulations | | 45.2 | Qwen3-1.7B (OPD+Clip) | 2026-07-15 |
| On the Step Length Confounding in LLM Reasoning Data Selection | ✓ Link | 45.0 | Qwen3-8B-Base (Aslec-casl) | 2026-04-08 |
| MHPO: Modulated Hazard-aware Policy Optimization for Stable Reinforcement Learning | | 44.7 | MHPO (Qwen3-4B-Base) | 2026-03-14 |
| LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language Models | ✓ Link | 44.67 | LEAD (DeepSeek-R1-Distill-Qwen-7B) | 2026-05-10 |
| | 44.6 | Openai-o1-preview | |
| Group Adaptive Clipping Policy Optimization | ✓ Link | 44 | DeepSeek-R1-Distill-Qwen-1.5B + GAPO | 2026-08-31 |
| Recovering Reliable Supervision via Critique-Guided On-Policy Distillation | ✓ Link | 43.44 | CriticOPD (Olmo-3-7B-Instruct-SFT) | 2026-10-14 |
| Composition-RL: Compose Your Verifiable Prompts for Reinforcement Learning of Large Language Models | ✓ Link | 43.3 | Qwen3-14B-Base + Composition-RL (12K subset) | 2026-02-12 |
| Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability | | 43.02 | Strict top-16 (Qwen3-235B-A22B-Instruct-2507 -> Granite-4.1-8B) | 2026-10-06 |
| Decoupling Reasoning and Confidence: Resurrecting Calibration in Reinforcement Learning from Verifiable Rewards | ✓ Link | 41.6 | Qwen3-8B (DCPO) | 2026-03-10 |
| PACED: Distillation at the Frontier of Student Competence | | 41.6 | Paced Forward KL (Qwen3-8B) | 2026-03-11 |
| Unlocking Token Rewards via Training-Free Reward Attribution | | 40.3 | Qwen2.5-Math-7B + GRPO (P2T token reward) (extra data) | 2026-06-03 |
| Qwen3 Technical Report | ✓ Link | 40.1 | Qwen3-235B-A22B (Non-thinking) | 2025-05-14 |
| BeamPERL: Parameter-Efficient RL with Verifiable Rewards Specializes Compact LLMs for Structured Beam Mechanics Reasoning | ✓ Link | 40.0 | BeamPERL (best-performing checkpoint) | 2026-03-04 |
| On the Off-Policy Teacher in On-Policy Distillation | | 39.8 | SCOUT | 2026-09-29 |
| DeepSeek-V3 Technical Report | ✓ Link | 39.2 | DeepSeek-V3 | 2024-12-27 |
| Cut Your Losses! Learning to Prune Paths Early for Efficient Parallel Reasoning | ✓ Link | 37.92 | STOP (DS-Qwen-2.5-1.5B) | 2026-04-17 |
| PRISM: Demystifying Retention and Interaction in Mid-Training | | 37.18 | Granite-3.3 (8B) PRISM | 2026-03-17 |
| On-policy Distillation with Verifiable Reward | ✓ Link | 36.9 | OPDVR (Qwen3-4B) | 2026-08-25 |
| Sparse but Critical: A Token-Level Analysis of Distributional Shifts in RLVR Fine-Tuning of LLMs | | 36.74 | Qwen2.5-Math-7B with DAPO (High-KL boost) | 2026-03-23 |
| Rectifying LLM Thought from Lens of Optimization | ✓ Link | 36.3 | DeepSeek-R1-Distill-Qwen-1.5B with PPO+RePro | 2025-12-01 |
| Towards Full Pipeline FP8 Reinforcement Learning for LLMs | | 33.5 | Qwen3-8B-Base + Calibrated Clipping (Blockwise FP8) | 2026-09-19 |
| From Imitation to Discrimination: Toward A Generalized Curriculum Advantage Mechanism Enhancing Cross-Domain Reasoning Tasks | | 33.3 | RLOO (with CAPO) (Qwen2.5-7B-Math) | 2025-12-02 |
| Beyond Imitation: Filtering On-Policy Distillation by Reasoning Progress | | 32.5 | R²-OPD (DeepSeek-R1-Distill-Qwen-1.5B) | 2026-08-19 |
| MixSD: Mixed Contextual Self-Distillation for Knowledge Injection | | 32.1 | Qwen3-8B with MixSD (lambda=0.7) | 2026-05-16 |
| ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy Learning | ✓ Link | 32.1 | ReSPO (Qwen3-30B-A3B-Base, N=16) | 2026-09-28 |
| ASI-Evolve: AI Accelerates AI | ✓ Link | 31.67 | Algorithm1 (ASI-Evolve) | 2026-03-31 |
| When Reasoning Meets Its Laws | ✓ Link | 31.3 | DeepSeek-R1-8B (SFT-Compo) | 2025-12-19 |
| Pushing the Boundaries of Natural Reasoning: Interleaved Bonus from Formal-Logic Verification | | 30.2 | FLV-RL (14B) | 2026-01-30 |
| SPARK: Stepwise Process-Aware Rewards for Reference-Free Reinforcement Learning | | 30.0 | PRM-CoT (Process-Aware) (Qwen2.5-Math-7B) | 2025-12-02 |
| LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models | | 30.00 | LLaDA MoE v2 Instruct | 2026-08-04 |
| KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization | | 29.7 | KV4-RL (Qwen3-8B-Base) | 2026-10-10 |
| Understanding Evolution Strategies for LLM Reasoning: Broader Reasoning Coverage than GRPO | | 29.0 | DeepSeek-R1-Distill-Qwen-1.5B (ES to GRPO) | 2026-08-27 |
| SCALER:Synthetic Scalable Adaptive Learning Environment for Reasoning | ✓ Link | 27.29 | Qwen3-4B-base + SCALER | 2026-01-08 |
| WHALE: A Simple Recipe for Joint Harness-Weight Optimization | ✓ Link | 27.08 | Qwen3.5-2B + WHALE (E0.6,I6) | 2026-08-31 |
| DenoiseRL: Bootstrapping Reasoning Models to Recover from Noisy Prefixes | ✓ Link | 27.0 | DenoiseRL-DAPO (Qwen3-8B-Base) | 2026-05-27 |
| Normalized Low-Rank Adaptation | ✓ Link | 26.7 | DS-R1-Distill-Qwen1.5B NoRA | 2026-08-31 |
| FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning | ✓ Link | 26.6700 | Phi-4-Reasoning (FROST) | 2026-01-26 |
| nabla-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space | ✓ Link | 26.6 | ∇-Reasoner (Qwen-2.5-7B-Instruct) | 2026-03-05 |
| Branching On-Policy Distillation | | 26.04 | BranchOPD (extra data) | 2026-10-09 |
| TIP: Token Importance in On-Policy Distillation | ✓ Link | 26.0 | TIP Soft-OR 50% (Llama-3.3-70B → Llama-3.1-8B) | 2026-04-15 |
| Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training | | 25.2 | Qwen3-1.7B (student) via two-stage bridge from RL-trained Qwen3-8B teacher | 2026-05-12 |
| Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO | ✓ Link | 24.4 | 4B→14B S2L | 2026-06-02 |
| Expert-Space Exploration in MoE Reinforcement Learning | ✓ Link | 24.4 | Qwen3-30B-A3B + ESRL | 2026-09-11 |
| Can LLMs Learn to Reason Robustly under Noisy Supervision? | ✓ Link | 23.8 | Qwen3-4B-Base (with OLR) | 2026-04-05 |
| Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance | ✓ Link | 23.75 | Falcon-H1-34B-Instruct | 2025-07-30 |
| Qwen2.5 Technical Report | ✓ Link | 23.3 | Qwen2.5-72B-Instruct | 2024-12-19 |
| Claude 3.7 Sonnet and Claude Code | | 23.3 | Claude 3.7 Sonnet | 2025-02-24 |
| Open-MOPD: Diagnosing and Fixing Capability Imbalance in Multi-Teacher On-Policy Distillation | ✓ Link | 21.98 | Open-MOPD (SmolLM3-3B) | 2026-08-19 |
| Tool Verification for Test-Time Reinforcement Learning | ✓ Link | 21.7 | T³RL (Qwen2.5-Math-1.5B, 7B verifier) | 2026-03-02 |
| DiRL: An Efficient Post-Training Framework for Diffusion Language Models | ✓ Link | 21.5 | DiRL-8B-Instruct | 2025-12-23 |
| Residual Context Diffusion Language Models | ✓ Link | 21.46 | SDAR (8B, b=32) with RCD | 2026-01-30 |
| The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents | | 20.8 | Qwen2.5-3B with CAR (MSCR) | 2026-01-12 |
| Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge | ✓ Link | 20.6 | Multiplex Thinking (DeepSeek-R1-Distill-Qwen-7B) | 2026-01-13 |
| Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance | ✓ Link | 20.41 | FEST-DPO | 2026-05-14 |
| PretrainZero: Reinforcement Active Pretraining | | 20.0 | PretrainZero (Qwen3-8B-Base) | 2025-12-03 |
| Adaptive Ability Decomposing for Unlocking Large Reasoning Model Effective Reinforcement Learning | | 20.0 | A2D (Qwen2.5-7B-Instruct) | 2026-01-31 |
| Stratagem: Learning Transferable Reasoning via Trajectory-Modulated Game Self-Play | ✓ Link | 20.0 | Stratagem (All Games) | 2026-04-20 |
| Beyond Policy Rollouts: Embedding-Space Directional Exploration with Anchored Noise for LLM RLVR | | 20.0 | EDEN (T=1.4, Qwen2.5-Math-1.5B) | 2026-10-12 |
| Can LLMs Guide Their Own Exploration? Gradient-Guided Reinforcement Learning for LLM Reasoning | | 19.9 | G2RL (Qwen3-4B-Base) | 2025-12-17 |
| Long Live The Balance: Information Bottleneck Driven Tree-based Policy Optimization | ✓ Link | 19.7 | IBTPO (Qwen3-8B-Base) | 2026-05-27 |
| J-Zero: Unified Challenger--Solver--Judge Co-Evolution from Zero Data | ✓ Link | 19.58 | J-Zero-8B | 2026-08-27 |
| Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration | ✓ Link | 17.9 | Qwen2.5-14B-Instruct with NExt | 2026-04-13 |
| SPOT: Sparse Probing and Outcome Calibration for On-Policy Distillation | ✓ Link | 17.5 | SPOT Qwen3-4B-Base | 2026-08-05 |
| Think Anywhere in Code Generation | ✓ Link | 17.3 | Think-Anywhere | 2026-03-31 |
| Learning to Hint for Reinforcement Learning | ✓ Link | 16.9 | HiLL (Qwen2.5-7B-Instruct) | 2026-04-01 |
| Stabilizing Reinforcement Learning for Diffusion Language Models | ✓ Link | 16.7 | StableDRL (SDAR-8B, static sampling) | 2026-03-06 |
| Harnessing LLM Agents with Skill Programs | | 16.7 | HASP-Evolve + RS (7B-Inst) | 2026-05-18 |
| ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection | ✓ Link | 16.56 | ProFit (Qwen3-14B-Base) | 2026-01-14 |
| Learn What's Left, Not What's Mastered: Saturation Aware Advantage Reweighting for Multi-Reward Policy Optimization | | 16.5 | Qwen2.5-7B-Instruct with SA-MRPO (3 objectives) (extra data) | 2026-08-17 |
| Dynamic Important Example Mining for Reinforcement Finetuning | ✓ Link | 16.2 | Qwen3-4B + DIEM | 2026-08-29 |
| | 16 | Claude3.5-Sonnet | |
| Off-the-Shelf LLMs as Process Scorers: Training-Free Alternative to PRMs for Mathematical Reasoning | | 15.6 | CGS (Qwen2.5-7B → Qwen2.5-72B) | 2026-06-01 |
| Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVR | ✓ Link | 14.0 | DATPO (Qwen3-4B-Base) | 2026-09-08 |
| OVD: On-policy Verbal Distillation | | 13.3 | OVD (Qwen2.5-Math-1.5B, 2 random examples, 800 steps) | 2026-01-29 |
| Trust-Region Behavior Blending for On-Policy Distillation | | 10.2 | Trust-Region behavior Blending (Qwen3-1.7B-Base ← Qwen3-8B) | 2026-05-29 |
| Enabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL Regularization | ✓ Link | 10.07 | Proximal KL (Qwen3-1.7B-Base, on-policy, lr 5e-6) | 2026-10-13 |
| The Molecular Structure of Thought: Mapping the Topology of Long Chain-of-Thought Reasoning | | 10.0 | Llama-3.1-8B-Instruct + 35K QwQ-Mole-Syn + RL | 2026-01-09 |
| Coupled Variational Reinforcement Learning for Language Model General Reasoning | ✓ Link | 9.5 | Qwen3-14B with CoVRL | 2025-12-14 |
| Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models | | 9.5 | WeDLM-8B (IMH) | 2026-04-01 |
| Gains and Collapse in On-Policy Distillation:A Reinforcement Learning Perspective | ✓ Link | 9.17 | 4B → SFT (warmup) | 2026-10-02 |
| LastOPD: Taming Collapse in Latent On-Policy Distillation | ✓ Link | 7.92 | LastOPD (Qwen3-8B -> Qwen3-1.7B-Base) | 2026-09-23 |
| Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models | | 2.64 | Llama 3.2–3B–Instruct (mid-trained, n=32) | 2026-05-08 |