| Efficient Agentic Reasoning Through Self-Regulated Simulative Planning | ✓ Link | 99.2 | SR 2 AM-v1.0-30B | 2026-05-21 |
| Announcing Bonsai 27B: The First 27B-Class Model to Run on a Phone | ✓ Link | 99.20 | Ternary Bonsai 27B | 2026-07-14 |
| Unlocking Lossless Speedups in LLMs via Discrete Diffusion | ✓ Link | 98.9 | Uno (K2-Horizon-7B-Uno) | 2026-09-03 |
| Learning More by Training Less: Uncertainty-Guided Progressive Contraction for Efficient Reasoning Post-Training | ✓ Link | 98.6 | Qwen3.5-9B (rho=30%) | 2026-10-09 |
| GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models | ✓ Link | 98.2 | GLM-4.5 (355B) | 2025-08-08 |
| GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models | ✓ Link | 98.1 | GLM-4.5-Air | 2025-08-08 |
| Qwen3 Technical Report | ✓ Link | 98.0 | Qwen3-235B-A22B (Thinking) | 2025-05-14 |
| JudgeRLVR: Judge First, Generate Second for Efficient Reasoning | | 98.0 | Vanilla RLVR (Qwen3-30B-A3B) | 2026-01-13 |
| Introspective Diffusion Language Models | ✓ Link | 97.6 | I-DLM (32B) | 2026-04-13 |
| CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning | ✓ Link | 97.6 | CopT (Qwen3-8B) | 2026-05-19 |
| Kimi K2: Open Agentic Intelligence | ✓ Link | 97.4 | Kimi K2 | 2025-07-28 |
| Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling | ✓ Link | 97.4 | Falcon-H1R (7B) | 2026-01-05 |
| ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning | ✓ Link | 97.4 | ThoughtFold (Qwen3-8B) | 2026-06-02 |
| DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | ✓ Link | 97.3 | DeepSeek-R1 | 2025-01-22 |
| Recursive Think-Answer Process for LLMs and VLMs | | 97.3 | MiMo-RL-7B-R-TAP | 2026-03-02 |
| One Model to Loop Them All: Elastic Latent Recurrence for Language Models | | 97.2 | Koi-4B | 2026-10-08 |
| ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search | ✓ Link | 97.13 | ZGCM-1 | 2026-09-11 |
| On-Policy Self-Distillation without Any Supervision | ✓ Link | 97.0 | u-OPSD Qwen3-30B-A3B-Instruct-2507 | 2026-08-09 |
| A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM | ✓ Link | 97.0 | A*-Thought-V2 (Qwen3.5-9B, tau=60) | 2026-09-07 |
| Rethinking Sample Polarity in Reinforcement Learning with Verifiable Rewards | | 96.9 | A3PO (Deepseek-R1-Distill-Qwen-7B) | 2025-12-25 |
| Scaling Embeddings Outperforms Scaling Experts in Language Models | | 96.8 | LongCat-Flash-Lite (68.5B) | 2026-01-29 |
| iGRPO: Self-Feedback-Driven LLM Reasoning | | 96.7 | OpenMath-Nemotron-14B + iGRPO | 2026-02-09 |
| ERNIE 4.5 Technical Report | ✓ Link | 96.4 | ERNIE-4.5-300B-A47B | 2025-06-30 |
| LongCat-Flash Technical Report | ✓ Link | 96.40 | LongCat-Flash | 2025-09-01 |
| SFT Conflicts, RL Coexists: A Theoretical and Empirical Analysis of Multi-Task Learning for LLMs | ✓ Link | 96.2 | Adapted Parallel-RL (DeepSeek-R1-Distill-Qwen-7B) | 2026-08-06 |
| LFM2.5-24B-A2B: Built for Fast On-Device Agents | | 96.04 | LFM2.5-24B-A2B | 2026-10-13 |
| CoDiQ: Test-Time Scaling for Controllable Difficult Question Generation | ✓ Link | 96.0 | CoDiQ-L3-4B | 2026-02-02 |
| EXAONE 4.0: Unified Large Language Models Integrating Non-reasoning and Reasoning Modes | ✓ Link | 95.8 | EXAONE 4.0 32B (Reasoning) | 2025-07-15 |
| Calibrating Teacher--Student Discrepancy for On-Policy Distillation | | 95.8 | Cal-OPD (Qwen3-30B-A3B-Thinking-2507 -> Qwen3-4B) | 2026-09-18 |
| Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models | ✓ Link | 95.5 | Nanbeige4-3B-Base + Ring-Lite SFT (extra data) | 2025-12-06 |
| Self-Hinting Language Models Enhance Reinforcement Learning | ✓ Link | 95.4 | SAGE (Qwen3-4B-Instruct) | 2026-02-03 |
| Every Coin Has Two Sides: On the Dual Nature of Generalization in On-Policy Distillation of Large Language Models | | 95.4 | DS-R1-Distill-Qwen-7B + OPD Dynamic Sampling | 2026-08-17 |
| TGRL: Temperature-Grouped Reinforcement Learning for Efficient Exploration in LLMs | ✓ Link | 95.3 | TGRL (Qwen3-14B) (extra data) | 2026-09-27 |
| Efficient Reasoning with Balanced Thinking | ✓ Link | 95.2 | ReBalance (QwQ-32B) | 2026-03-12 |
| Post-Trained MoE Can Skip Half Experts via Self-Distillation | ✓ Link | 95.2 | ZEDA (Qwen3-30B-A3B) | 2026-05-18 |
| Agentic Chain-of-Thought Steering for Efficient and Controllable LLM Reasoning | ✓ Link | 95.2 | ACTS (Qwen3-8B) | 2026-06-02 |
| Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning | | 95.1 | ERC-DAPO (DeepSeek-R1-Distill-Qwen-7B) | 2025-12-05 |
| Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability | ✓ Link | 95.1 | Qwen3-14B (Math-CoT SFT) | 2026-04-08 |
| Figure It Out: Improving the Frontier of Reasoning with Active Visual Thinking | ✓ Link | 95.0 | FigR | 2025-12-30 |
| Efficient Reasoning on the Edge | | 95.0 | Qwen2.5-7B-Instruct (OT3, dense) | 2026-03-17 |
| Revisiting Complete Reasoning Traces for Post-Training | ✓ Link | 95 | Qwen2.5-32B-Instruct + segment SFT (s1K-1.1, Prefix) | 2026-09-07 |
| Heterogeneous Agent Collaborative Reinforcement Learning | | 94.8 | Qwen3-4B-Instruct (HACPO) | 2026-03-03 |
| Distilling Long-CoT Reasoning through Collaborative Step-wise Multi-Teacher Decoding | ✓ Link | 94.8 | R1-Qwen (32B) with CoRD (heterogeneous teachers) | 2026-05-04 |
| Filter, Then Reweight: Rethinking Optimization Granularity in On-Policy Distillation | ✓ Link | 94.69 | FiRe-OPD | 2026-06-01 |
| OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization | ✓ Link | 94.66 | OSCAR (GLM-4.7, 358B) | 2026-05-18 |
| Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math | | 94.6 | Phi-4-Mini-Reasoning | 2025-04-30 |
| Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment | ✓ Link | 94.6 | Qwen-3-14B (fine-tuned with Rank-Surprisal Ratio selection) | 2026-01-20 |
| MiniCPM5-2B | ✓ Link | 94.6 | MiniCPM5-2B | 2026-09-07 |
| Co-Evolving Policy Distillation | | 94.5 | CoPD | 2026-04-29 |
| Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models | ✓ Link | 94.2 | Qwen3-30B-A3B-Thinking with PUMA | 2026-05-17 |
| JustRL: Scaling a 1.5B LLM with a Simple RL Recipe | ✓ Link | 94.15 | JustRL-Nemotron 1.5B | 2025-12-18 |
| InftyThink+: Effective and Efficient Infinite-Horizon Reasoning via Reinforcement Learning | ✓ Link | 94.09 | InftyThink+ (Qwen3-4B-Base, task-only RL) | 2026-02-06 |
| PACED: Distillation at the Frontier of Student Competence | | 94.0 | Paced Forward KL (Qwen3-8B) | 2026-03-11 |
| HyperCLOVA X THINK Technical Report | ✓ Link | 93.8 | HyperCLOVA X SEED 14B Think | 2025-06-27 |
| Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models | ✓ Link | 93.7 | Youtu-LLM 2B | 2025-12-31 |
| Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning | ✓ Link | 93.6 | NPR (Qwen3-4B-Instruct-2507) | 2025-12-08 |
| FlowBalance: Verifier-Grounded Self-Improvement from On-Policy Reasoning Experience | ✓ Link | 93.52 | Qwen3-8B + FlowBalance | 2026-09-03 |
| IIB-LPO: Latent Policy Optimization via Iterative Information Bottleneck | | 93.5 | I2B-LPO (Qwen3-14B) | 2026-01-09 |
| Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models | | 93.4 | MELT-1.6B | 2026-05-08 |
| Efficient Reasoning Training Does Not Always Harm CoT Faithfulness and Monitorability | ✓ Link | 93.0 | ThinkPrune (Qwen3-4B) | 2026-10-02 |
| ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning | ✓ Link | 92.7 | ResRL (Qwen3-8B) | 2026-05-01 |
| The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping | ✓ Link | 92.51 | Qwen3-8B + MEDS | 2026-04-13 |
| Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation | ✓ Link | 92.4 | OPDLM-MATH-8B-Thinking | 2026-06-04 |
| TPPO: Tracking-Value PPO For Language Model Reasoning | ✓ Link | 92.3 | TPPO (Qwen3-4B) | 2026-10-12 |
| LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language Models | ✓ Link | 92.27 | LEAD (DeepSeek-R1-Distill-Qwen-7B) | 2026-05-10 |
| GRAIL: Gradient-Reweighted Advantages for Reinforcement Learning with Verifiable Rewards | ✓ Link | 92.07 | GRAIL (Qwen3-8B) | 2026-06-03 |
| On-Policy Distillation with Negative-Policy Rollouts | ✓ Link | 91.8 | Qwen3-8B + NP-OPD (OPD 2, alpha=0.5, thinking mode) | 2026-10-06 |
| MiniCPM5-1B | ✓ Link | 91.6 | MiniCPM5-1B (Thinking) | 2026-05-25 |
| TRIAGE: Direction-Aware Mismatch Stabilization of Native NVFP4 Reinforcement Learning | ✓ Link | 91.6 | Qwen3-30B-A3B-Base (NVFP4+TRIAGE, step 1700) | 2026-10-05 |
| ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models | ✓ Link | 91.4 | ThreadWeaver (Qwen3-8B) | 2025-11-24 |
| LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning | ✓ Link | 91.4 | LYNX (QwQ-32B) | 2025-12-05 |
| Qwen3 Technical Report | ✓ Link | 91.2 | Qwen3-235B-A22B (Non-thinking) | 2025-05-14 |
| Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs? | ✓ Link | 90.93 | DeepSeek-R1-Distill-Qwen-7B (SFT on D_ug) | 2026-03-25 |
| TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning | | 90.7 | TERMINATOR (Qwen3-8B) | 2026-03-13 |
| Towards Full Pipeline FP8 Reinforcement Learning for LLMs | | 90.7 | Qwen3-8B-Base + Calibrated Clipping (Blockwise FP8) | 2026-09-19 |
| Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs | ✓ Link | 90.4 | Phi-4-Mini reasoning | 2025-02-24 |
| From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Space | ✓ Link | 90.31 | DSRL (Qwen3-8B) | 2026-04-15 |
| DeepSeek-V3 Technical Report | ✓ Link | 90.2 | DeepSeek-V3 | 2024-12-27 |
| On the Off-Policy Teacher in On-Policy Distillation | | 90.0 | SCOUT | 2026-09-29 |
| Recovering Reliable Supervision via Critique-Guided On-Policy Distillation | ✓ Link | 89.08 | CriticOPD (Olmo-3-7B-Instruct-SFT) | 2026-10-14 |
| LFM2.5-8B-A1B: an Even Better on-Device Mixture-of-Experts | ✓ Link | 88.76 | LFM2.5-8B-A1B | 2026-05-28 |
| DenoiseRL: Bootstrapping Reasoning Models to Recover from Noisy Prefixes | ✓ Link | 88.2 | DenoiseRL-DAPO (Qwen3-8B-Base) | 2026-05-27 |
| Agentic Critical Training | | 87.73 | ACT (Qwen3-8B) | 2026-03-09 |
| Rectifying LLM Thought from Lens of Optimization | ✓ Link | 87.7 | DeepSeek-R1-Distill-Qwen-1.5B with PPO+RePro | 2025-12-01 |
| KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization | | 87.7 | KV4-RL (Qwen3-8B-Base) | 2026-10-10 |
| LFM-1B-Math: Can Small Models Be Concise Reasoners? | | 87.48 | LFM-1.3B-Distill (32K output budget) | 2025-06-24 |
| ESPO: Early-Stopping Proximal Policy Optimization | | 87.42 | ESPO (DeepSeek-R1-Distill-Qwen-7B) | 2026-05-28 |
| NGM: A Plug-and-Play Training-Free Memory Module for LLMs | ✓ Link | 87.0 | Qwen3-14B + NGM | 2026-05-16 |
| Selection Is Not Learning: Choosing Model Futures | | 86.87 | EvoHeir (Gemma-4-E4B-it) | 2026-10-15 |
| On-Policy Self-Distillation for Reasoning Compression | ✓ Link | 86.6 | Qwen3-8B (OPSDC) | 2026-03-05 |
| Unsupervised Process Reward Models | | 86.5 | uPRM | 2026-05-11 |
| ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy Learning | ✓ Link | 85.7 | ReSPO (Qwen3-30B-A3B-Base, N=16) | 2026-09-28 |
| ASI-Evolve: AI Accelerates AI | ✓ Link | 85.6 | Algorithm1 (ASI-Evolve) | 2026-03-31 |
| SpecFold: Folding Multi-Branch Redundancy for Faster Speculative Decoding in Diffusion Language Models | ✓ Link | 85.6 | SpecFold (Nemotron-Labs-Diffusion-14B) | 2026-10-06 |
| Group Adaptive Clipping Policy Optimization | ✓ Link | 85.5 | DeepSeek-R1-Distill-Qwen-1.5B + GAPO | 2026-08-31 |
| SPARK: Stepwise Process-Aware Rewards for Reference-Free Reinforcement Learning | | 85.4 | PRM-CoT (Process-Aware) (Qwen2.5-Math-7B) | 2025-12-02 |
| MixSD: Mixed Contextual Self-Distillation for Knowledge Injection | | 85.4 | Qwen3-8B with MixSD (lambda=0.7) | 2026-05-16 |
| Can LLMs Learn to Reason Robustly under Noisy Supervision? | ✓ Link | 85.2 | Qwen3-4B-Base (with OLR) | 2026-04-05 |
| DiRL: An Efficient Post-Training Framework for Diffusion Language Models | ✓ Link | 85.1 | DiRL-8B-Instruct | 2025-12-23 |
| Beyond the Stability-Exploration Dilemma: Environmental Regularization for LLM Policy Optimization | | 85.0 | ERPO (Qwen2.5-32B) | 2026-08-24 |
| RISE: Recursive Improvement via Self-Extrapolating Policy Distillation | | 84.9 | RISE (weight,Qwen3-8B,3-seedmean) | 2026-09-04 |
| dots.llm1 Technical Report | ✓ Link | 84.8 | dots.llm1.inst | 2025-06-06 |
| On-policy Distillation with Verifiable Reward | ✓ Link | 84.7 | OPDVR (Qwen3-4B) | 2026-08-25 |
| OmniOPD: Logit-Free On-Policy Distillation via Speculative Verification | | 84.61 | OmniOPD (Qwen3-1.7B student, Qwen3-32B teacher) | 2026-05-31 |
| TraPO: A Semi-Supervised Reinforcement Learning Framework for Boosting LLM Reasoning | ✓ Link | 84.6 | Qwen2.5-Math-7B with TraPO (4K labels, 12K unlabeled) | 2025-12-15 |
| Unlocking Token Rewards via Training-Free Reward Attribution | | 84.5 | Qwen2.5-Math-7B + GRPO (P2T token reward) (extra data) | 2026-06-03 |
| Understanding Evolution Strategies for LLM Reasoning: Broader Reasoning Coverage than GRPO | | 84.5 | DeepSeek-R1-Distill-Qwen-1.5B (GRPO to ES) | 2026-08-27 |
| Normalized Low-Rank Adaptation | ✓ Link | 84.5 | DS-R1-Distill-Qwen1.5B NoRA | 2026-08-31 |
| SCALER:Synthetic Scalable Adaptive Learning Environment for Reasoning | ✓ Link | 84.4 | Qwen3-4B-base + SCALER | 2026-01-08 |
| Beyond Stochastic Exploration: What Makes Training Data Valuable for Agentic Search | ✓ Link | 84.2 | GRPO + HiExp (Qwen2.5-Math-7B) | 2026-04-09 |
| Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance | ✓ Link | 83.80 | Falcon-H1-34B-Instruct | 2025-07-30 |
| J-Zero: Unified Challenger--Solver--Judge Co-Evolution from Zero Data | ✓ Link | 83.4 | J-Zero-8B | 2026-08-27 |
| Long Live The Balance: Information Bottleneck Driven Tree-based Policy Optimization | ✓ Link | 83.3 | IBTPO (Qwen3-8B-Base) | 2026-05-27 |
| When Reasoning Meets Its Laws | ✓ Link | 83.0 | DeepSeek-R1-8B (SFT-Compo) | 2025-12-19 |
| ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection | ✓ Link | 82.85 | ProFit (Qwen3-14B-Base) | 2026-01-14 |
| Your Group-Relative Advantage Is Biased | | 82.8 | DAPO+HA-DW (Qwen-3-8B-Base) | 2026-01-13 |
| Solving Physics Olympiad via Reinforcement Learning on Physics Simulators | ✓ Link | 82.8 | Qwen2.5-32B + RL (synthetic) | 2026-04-13 |
| PRISM: Demystifying Retention and Interaction in Mid-Training | | 82.73 | Granite-4-H Micro (3B) PRISM | 2026-03-17 |
| Guided Self-Evolving LLMs with Minimal Human Supervision | | 82.6 | Qwen3-8B-Base (R-Few 5%) | 2025-12-02 |
| Enabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL Regularization | ✓ Link | 82.4 | Proximal KL (Qwen3-4B-Base, off-policy, staleness 4, lr 5e-6) | 2026-10-13 |
| Claude 3.7 Sonnet and Claude Code | | 82.2 | Claude 3.7 Sonnet | 2025-02-24 |
| Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes | ✓ Link | 82.0 | Local Support Matching (Qwen2.5-7B-Instruct, unmasked) | 2026-03-26 |
| PretrainZero: Reinforcement Active Pretraining | | 81.9 | PretrainZero (Qwen3-8B-Base) | 2025-12-03 |
| Learning to Hint for Reinforcement Learning | ✓ Link | 81.8 | HiLL (Qwen2.5-7B-Instruct) | 2026-04-01 |
| Sharpen Without Search: On-Policy Distillation of Sequence-Level Power Distribution | ✓ Link | 81.25 | OPPD (Qwen2.5-Math-7B, seed 2) | 2026-10-05 |
| Branching On-Policy Distillation | | 81.24 | BranchOPD (extra data) | 2026-10-09 |
| Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance | ✓ Link | 81.1 | FEST-GRPO | 2026-05-14 |
| HalluGuard: Demystifying Data-Driven and Reasoning-Driven Hallucinations in LLMs | | 81.0 | HalluGuard (Qwen2.5-Math-7B) | 2026-01-26 |
| APPO: Agentic Procedural Policy Optimization | ✓ Link | 81.0 | APPO (Qwen2.5-7B-Instruct) | 2026-06-10 |
| Can LLMs Guide Their Own Exploration? Gradient-Guided Reinforcement Learning for LLM Reasoning | | 80.8 | G2RL (Qwen3-4B-Base) | 2025-12-17 |
| ALoDLM: Adaptively Looped Diffusion Language Models | ✓ Link | 80.8 | ALoDLM-8B | 2026-10-03 |
| nabla-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space | ✓ Link | 80.4 | ∇-Reasoner (Qwen-2.5-7B-Instruct) | 2026-03-05 |
| Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs | | 80.4 | DeepSeek-R1-Distill-Qwen-1.5B (Graph-based CoT Pruning) | 2026-04-07 |
| Guiding LLM Post-training Data Engineering with Model Internals from Sparse Autoencoders | | 80.3 | Qwen2.5-Math-7B (SaeRL, GRPO) | 2026-05-26 |
| SPOT: Sparse Probing and Outcome Calibration for On-Policy Distillation | ✓ Link | 79.88 | SPOT Qwen3-4B-Base | 2026-08-05 |
| Label-free steering: Compressing test-time reinforcement learning into bias-only subspaces | ✓ Link | 79.5 | Qwen2.5-Math-7B + Label-Free Bias-Only TTRL | 2026-09-28 |
| Demystifying Hidden-State Recurrence: Switchable Latent Reasoning with On-Policy Reinforcement Learning | | 79.3 | Switch (ours, after Switch-GRPO) | 2026-06-11 |
| Stabilizing Reinforcement Learning for Diffusion Language Models | ✓ Link | 79.2 | StableDRL (SDAR-8B, static sampling) | 2026-03-06 |
| Multi-Block Diffusion Language Models | ✓ Link | 79.2 | MBD-LLaDA2-Mini | 2026-06-30 |
| TIP: Token Importance in On-Policy Distillation | ✓ Link | 79.1 | TIP Soft-OR 50% (Qwen3-8B → Qwen3-4B) | 2026-04-15 |
| Length-Unbiased Sequence Policy Optimization: Revealing and Controlling Response Length Variation in RLVR | ✓ Link | 78.4 | Qwen2.5-7B-Base (LUSPO) | 2026-02-05 |
| Geometry Conflict: Explaining and Controlling Forgetting in LLM Continual Post-Training | ✓ Link | 78.2 | Qwen3-14B with GCWM | 2026-05-10 |
| Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge | ✓ Link | 78.0 | Multiplex Thinking (DeepSeek-R1-Distill-Qwen-7B) | 2026-01-13 |
| CLIPO: Contrastive Learning in Policy Optimization Generalizes RLVR | ✓ Link | 77.87 | GMPO + CLIPO | 2026-03-10 |
| The Molecular Structure of Thought: Mapping the Topology of Long Chain-of-Thought Reasoning | | 77.8 | Qwen-2.5-7B-Instruct + 20K OSS-Mole-Syn | 2026-01-09 |
| Residual Context Diffusion Language Models | ✓ Link | 77.6 | SDAR (8B, b=32) with RCD | 2026-01-30 |
| MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction | ✓ Link | 77.0 | MiniCPM-o 4.5-Instruct | 2026-04-30 |
| The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents | | 76.9 | Qwen2.5-3B with CAR (MSCR) | 2026-01-12 |
| From Imitation to Discrimination: Toward A Generalized Curriculum Advantage Mechanism Enhancing Cross-Domain Reasoning Tasks | | 76.8 | GRPO (with CAPO) (Qwen2.5-7B-Math) | 2025-12-02 |
| Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVR | ✓ Link | 76.4 | DATPO (Qwen3-4B-Base) | 2026-09-08 |
| Think Again or Think Longer? Selective Verification for Budget-Aware Reasoning | ✓ Link | 76.3 | Selective active verify (Qwen3-1.7B gate) | 2026-06-18 |
| Coupled Variational Reinforcement Learning for Language Model General Reasoning | ✓ Link | 76.1 | Qwen3-14B with CoVRL | 2025-12-14 |
| Representation-Space MMD for Diffusion Language Models | ✓ Link | 76.0 | DMax-Math-MMD | 2026-10-05 |
| Weak-Driven Learning: How Weak Agents make Strong Agents Stronger | ✓ Link | 75.4 | WMSS (Qwen3-8B-Base) | 2026-02-09 |
| DMax: Aggressive Parallel Decoding for dLLMs | ✓ Link | 75.4 | DMax-Math | 2026-04-09 |
| Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL | ✓ Link | 75.2 | Co-RL (Qwen2.5-7B, Different family) | 2026-08-19 |
| Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning | | 75.1 | Qwen2.5-Math-7B + ACE-DAPO | 2026-02-24 |
| daVinci-Env: Open SWE Environment Synthesis at Scale | ✓ Link | 72.6 | OpenSWE-72B (Qwen2.5-72B-Base) | 2026-03-13 |
| Reasoning Palette: Modulating Reasoning via Latent Contextualization for Controllable Exploration for (V)LMs | | 72.2 | Qwen3-8B-Base with Reasoning Palette (RLOO, Linear Decay) | 2025-12-19 |
| Rethinking RL for LLM Reasoning: It's Sparse Policy Selection, Not Capability Learning | ✓ Link | 70.6 | ReasonMaxxer (Qwen2.5-7B) | 2026-05-07 |
| Learning User Preferences Through Interaction for Long-Term Collaboration | ✓ Link | 70.35 | gpt-oss-20b + memory | 2026-01-06 |
| T3D: Few-Step Diffusion Language Models via Trajectory Self-Distillation with Direct Discriminative Optimization | ✓ Link | 70.0 | T3D (SDAR-4B-Chat) | 2026-02-12 |
| Trust-Region Behavior Blending for On-Policy Distillation | | 69.7 | Trust-Region behavior Blending (Qwen3-1.7B-Base ← Qwen3-8B) | 2026-05-29 |
| FlashREINFORCE: Critic-Free Single-Rollout Asynchronous RL for Agentic Language Models | ✓ Link | 69.2 | FlashREINFORCE (Qwen2.5-Math-1.5B) | 2026-09-14 |
| Learn What's Left, Not What's Mastered: Saturation Aware Advantage Reweighting for Multi-Reward Policy Optimization | | 67.7 | Qwen2.5-7B-Instruct with SA-MRPO (3 objectives) (extra data) | 2026-08-17 |
| HyperThink: Text-to-Parameter Hypernetworks for Efficient Reasoning | | 67.36 | HyperThink (SmolLM3-3B) (extra data) | 2026-10-02 |
| Dynamic Important Example Mining for Reinforcement Finetuning | ✓ Link | 65.9 | Qwen3-4B + DIEM | 2026-08-29 |
| SkillFactory: Self-Distillation For Learning Cognitive Behaviors | ✓ Link | 64.6 | Qwen2.5-7B-Instruct (SkillFactory with 1k SFT + GRPO) | 2025-12-03 |
| Multi-Stream LLMs: Unblocking Language Models with Parallel Streams of Thoughts, Inputs and Outputs | ✓ Link | 64.0 | Qwen3-4B (Stream) | 2026-05-12 |
| Agon: Competitive Cross-Model RL with Implicit Rival Grading of Reasoning | | 64.0 | Agon | 2026-07-08 |
| Group Sequence Policy Optimization | | 62.9 | AliceAI-T5-35B-A0.6B-Base | 2025-07-24 |
| Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO | ✓ Link | 62.7 | 4B→14B S2L | 2026-06-02 |
| FROST: Filtering Reasoning Outliers with Attention for Efficient Reasoning | ✓ Link | 59.8 | Phi-4-Reasoning (FROST) | 2026-01-26 |
| Composing What Each Teacher Learned: Multi-Teacher On-Policy Distillation through Teacher-Relative Shifts | | 59.4 | Δ-MOPD (1.5B student) | 2026-10-07 |
| LastOPD: Taming Collapse in Latent On-Policy Distillation | ✓ Link | 58.95 | LastOPD (Qwen3-4B -> Qwen3-1.7B-Base) | 2026-09-23 |
| NeuPAT: Neuron-aware Plasticity Allocation Tuning for Language-Preserving MLLMs | | 56.4 | NeuPAT (Qwen3-4B-Instruct-2507) (extra data) | 2026-08-08 |
| Rationale-Guided Policy Optimization: Learning to Reason with Adaptive Rationale Scaffolding | ✓ Link | 55.2 | RGPO | 2026-10-05 |
| Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models | | 54.0 | WeDLM-8B (IMH) | 2026-04-01 |
| From Gradients to Capabilities: Understanding Multi-Teacher On-Policy Distillation | | 53.8 | SmolLM3-3B (PG/domain-response SGD) | 2026-10-01 |
| Beyond Test-Time Training: Learning to Reason via Hardware-Efficient Optimal Control | ✓ Link | 52.8 | TTC-Net (Llama-3-Instruct-7B) | 2026-03-10 |
| Who Should Lead Decoding Now? Tracking Reliable Trajectories for Ensembling Masked Diffusion Language Models | | 48.6 | TIE (LLaDA + Dream; Token change count) | 2026-06-15 |
| Self-Play Search Distillation for Large Language Model Reasoning | | 46.8 | SPSD-Qwen3-4B-Base (OPSD) | 2026-09-25 |
| The Flexibility Trap: Why Arbitrary Order Limits Reasoning Potential in Diffusion Language Models | ✓ Link | 45.2 | JustGRPO (LLaDA 8B Instruct) | 2026-01-21 |
| SLPO: Scaling Latent Reasoning via a Surrogate Policy | ✓ Link | 45.0 | SLPO (Llama-3.2-3B-Instruct) | 2026-07-22 |
| NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction | ✓ Link | 43.91 | NCP-ArchPreview Stage-2 | 2026-09-09 |
| BitNet b1.58 2B4T Technical Report | ✓ Link | 43.40 | BitNet b1.58 2B4T | 2025-04-16 |
| Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models | | 38.11 | Llama 3.2–3B–Instruct (mid-trained, n=64) | 2026-05-08 |
| Pivot-SD: Efficient Self-Distillation for Masked Diffusion Language Models | ✓ Link | 37.47 | Pivot-SD | 2026-10-02 |
| C^2DLM: Causal Concept-Guided Diffusion Large Language Models | ✓ Link | 37.2 | C2DLM | 2025-11-27 |
| Data Science and Technology Towards AGI Part I: Tiered Data Management | ✓ Link | 37.02 | MiniCPM-1.2B (Math-L3) | 2026-02-09 |
| Beyond Hard Masks: Progressive Token Evolution for Diffusion Language Models | ✓ Link | 36.33 | EvoToken-DLM (LLaDA-Instruct-8B, 10k FT) | 2026-01-12 |
| dLLM: Simple Diffusion Language Modeling | ✓ Link | 35.40 | LLaDA-Instruct (reasoning SFT) | 2026-02-26 |
| Token-Level LLM Collaboration via FusionRoute | | 33.0 | FusionRoute (Llama-3.1-8B) | 2026-01-08 |