| Introducing GPT-5.2 | | 100.0 | GPT-5.2 | 2025-12-11 |
| Magenta: Closing the Loop Between Mathematical Reasoning and Lean Verification | | 100.0 | K2-Horizon-7B + Magenta | 2026-09-10 |
| Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation | | 98.6 | Nemotron-Cascade-2-30B-A3B | 2026-03-19 |
| Seed2.0 Model Card: Towards Intelligence Frontier for Real-World Complexity | | 98.3 | Seed2.0 Pro | 2026-06-30 |
| SimpleOPD: Simple Tokenizer-Agnostic On-Policy Distillation for Long-Context Reasoning | ✓ Link | 97.5 | Intern-S2-DS-OPD | 2026-08-14 |
| Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters | ✓ Link | 97.3 | Step-3.5-Flash | 2026-02-11 |
| Nemotron-Math: Efficient Long-Context Distillation of Mathematical Reasoning from Multi-Mode Supervision | ✓ Link | 97.29 | Qwen3-30B-A3B (fine-tuned on Nemotron-Math AoPS+StackExchange-Math subset, high reasoning, Python TIR) | 2025-12-17 |
| MAI-Thinking-1: Building a Hill-Climbing Machine | | 97.0 | MAI-Thinking-1 | 2026-06-02 |
| Kolibri: A Sovereign European Model on the Pareto Frontier | ✓ Link | 96.9 | Kolibri | 2026-10-03 |
| Falcon-H1R: Pushing the Reasoning Frontiers with a Hybrid Model for Efficient Test-Time Scaling | ✓ Link | 96.7 | Falcon-H1R (7B) with DeepConf@512 | 2026-01-05 |
| Long-horizon Reasoning Agent for Olympiad-Level Mathematical Problem Solving | | 96.6 | Intern-S1-MO | 2025-12-11 |
| Kimi K2.5: Visual Agentic Intelligence | ✓ Link | 96.1 | Kimi K2.5 | 2026-02-02 |
| DeepSeek-V3.2 Technical Report | ✓ Link | 96.0 | DeepSeek-V3.2-Speciale | 2025-12-02 |
| Streaming Communication in Multi-Agent Reasoning | ✓ Link | 95.83 | STREAM-MA (Claude Opus 4.6, Graph) | 2026-06-03 |
| Introducing Gemini 3 Flash | | 95.2 | Google Gemini 3 Flash (Thinking) | 2025-12-17 |
| A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLM | ✓ Link | 95.0 | A*-Thought-V2 (Qwen3.6-27B, tau=60) | 2026-09-07 |
| Introducing GPT-5 | | 94.6 | GPT-5 | 2025-08-07 |
| Achieving Gold-Medal-Level Olympiad Reasoning via Simple and Unified Scaling | ✓ Link | 94.6 | SU-01 | 2026-05-13 |
| MiMo-V2-Flash Technical Report | ✓ Link | 94.1 | MiMo-V2-Flash (Thinking) | 2026-01-06 |
| MOPD: Multi-Teacher On-Policy Distillation for Capability Integration in LLM Post-Training | | 94.1 | MiMo-V2-Flash + MOPD | 2026-06-29 |
| 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 |
| ReTerm: Recovery Training for Self-Improving Terminal Agents | ✓ Link | 93.3 | ReTerm-9B | 2026-10-13 |
| Intern-S1-Pro: Scientific Multimodal Foundation Model at Trillion Scale | | 93.1 | Intern-S1-Pro | 2026-03-26 |
| EXAONE 4.5 Technical Report | ✓ Link | 92.9 | EXAONE 4.5 (Reasoning) | 2026-04-09 |
| K-EXAONE Technical Report | ✓ Link | 92.8 | K-EXAONE (236B, reasoning) | 2026-01-05 |
| Introducing OpenAI o3 and o4-mini | ✓ Link | 92.7 | OpenAI o4-mini | 2025-04-16 |
| IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse | ✓ Link | 92.6 | DSA (30B) + IndexCache (1/4, searched pattern) | 2026-03-12 |
| Mach-Mind-4-Flash Technical Report | | 92.08 | Mach-Mind-4-Flash | 2026-07-10 |
| Group Entropy-Controlled Policy Optimization | | 91.4 | GEPO (Qwen3.5-9B) | 2026-07-18 |
| Unified Audio Intelligence Without Regressing on Text Intelligence | | 91.2 | Audex 30B-A3B | 2026-07-06 |
| Efficient Agentic Reasoning Through Self-Regulated Simulative Planning | ✓ Link | 91.1 | SR 2 AM-v1.0-30B | 2026-05-21 |
| Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning | | 91.0 | Ring-2.5-1T-Zero (High) | 2026-07-14 |
| Announcing Bonsai 27B: The First 27B-Class Model to Run on a Phone | ✓ Link | 90.84 | Ternary Bonsai 27B | 2026-07-14 |
| Unlocking Lossless Speedups in LLMs via Discrete Diffusion | ✓ Link | 90.7 | Uno (K2-Horizon-7B-Uno) | 2026-09-03 |
| Scalable Visual Pretraining for Language Intelligence | | 90.21 | Qwen3.5-35B-A3B + VP | 2026-07-10 |
| Limite 1B - Violetto: Technical Report | ✓ Link | 90.21 | Limite 1B - Violetto | 2026-10-06 |
| PRISM: Pushing the Frontier of Deep Think via Process Reward Model-Guided Inference | ✓ Link | 90.0 | PRISM (gpt-oss-20b; PRM-score Vote) | 2026-03-03 |
| ACC: Compiling Agent Trajectories for Long-Context Training | | 90.0 | Qwen3-30B-A3B-Thinking + ACC | 2026-05-21 |
| Thought-Level Beam Search for Reasoning | ✓ Link | 90.0 | Gambit (Qwen3-4B-Thinking-2507) | 2026-08-11 |
| Qwen3-VL Technical Report | ✓ Link | 89.7 | Qwen3-VL-235B-A22B-Thinking | 2025-11-26 |
| Granite 4.2 LLMs: How They're Built | ✓ Link | 89.17 | Granite-4.2-30B | 2026-08-25 |
| Recursive Think-Answer Process for LLMs and VLMs | | 88.7 | Qwen3-235B-A22B-R-TAP (MoE, 235B) | 2026-03-02 |
| Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities | | 88.0 | Gemini 2.5 Pro | 2025-07-07 |
| INTELLECT-3: Technical Report | ✓ Link | 88.0 | INTELLECT-3 | 2025-12-18 |
| Distribution-Aligned Sequence Distillation for Superior Long-CoT Reasoning | ✓ Link | 86.7 | DASD-30B-A3B-Thinking-Preview (30B MoE) | 2026-01-14 |
| Recursive Multi-Agent Systems | ✓ Link | 86.7 | RecursiveMAS | 2026-04-28 |
| SkillOS: Learning Skill Curation for Self-Evolving Agents | | 86.7 | SkillOS (Qwen3-8B curator) with Gemini-2.5-Pro executor | 2026-05-07 |
| Kwai Keye-VL-2.0 Technical Report | ✓ Link | 86.7 | Keye-VL-2.0-30B-A3B | 2026-06-09 |
| Data Science and Technology Towards AGI Part I: Tiered Data Management | ✓ Link | 86.5 | MiniCPM5-2B | 2026-02-09 |
| QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management | ✓ Link | 86.46 | QwenLong-L1.5-30B-A3B | 2025-12-15 |
| Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT Data | | 85.73 | Mixed SFT + RLVR | 2026-08-24 |
| Nanbeige4-3B Technical Report: Exploring the Frontier of Small Language Models | ✓ Link | 85.6 | Nanbeige4-3B-Thinking-2511 | 2025-12-06 |
| EXAONE 4.0: Unified Large Language Models Integrating Non-reasoning and Reasoning Modes | ✓ Link | 85.3 | EXAONE 4.0 32B (Reasoning) | 2025-07-15 |
| Solar Open Technical Report | ✓ Link | 84.3 | Solar Open (102B) | 2026-01-11 |
| Loop the Loopies! | | 83.75 | Loopie-20B-A2B (Thinking) | 2026-07-17 |
| Nemotron-Cascade: Scaling Cascaded Reinforcement Learning for General-Purpose Reasoning Models | | 83.3 | Nemotron-Cascade (14B, Thinking) | 2025-12-15 |
| PACT: From Credit Assignment to Critic Alignment | ✓ Link | 83.12 | PACT (Qwen3.5-4B) | 2026-09-22 |
| Qwen3-Coder-Next Technical Report | ✓ Link | 83.07 | Qwen3-Coder-Next | 2026-02-28 |
| Learning to Build the Environment: Self-Evolving Reasoning RL via Verifiable Environment Synthesis | | 83.0 | EvoEnv (Qwen3-4B-Thinking-2507) | 2026-05-14 |
| Nemotron 3 Nano Omni: Efficient and Open Multimodal Intelligence | ✓ Link | 82.1 | Nemotron 3 Nano Omni | 2026-04-27 |
| Qwen3 Technical Report | ✓ Link | 81.5 | Qwen3-235B-A22B (Thinking) | 2025-05-14 |
| CHIMERA: Compact Synthetic Data for Generalizable LLM Reasoning | | 80.7 | Qwen3-4B-Thinking-2507 (4B) fine-tuned on Chimera (SFT+RL) | 2026-03-01 |
| Safin-1: Safety from Within through Memory-Native State Evolution | ✓ Link | 80.1 | Safin-1-35B-A3B-SFT | 2026-08-31 |
| Introspective Diffusion Language Models | ✓ Link | 80.0 | I-DLM (32B) | 2026-04-13 |
| Stop When Reasoning Converges: Semantic-Preserving Early Exit for Reasoning Models | ✓ Link | 80.0 | Qwen3-30B-A3B-Thinking with PUMA | 2026-05-17 |
| JudgeRLVR: Judge First, Generate Second for Efficient Reasoning | | 79.7 | Mixed Strategy (Qwen3-30B-A3B) | 2026-01-13 |
| Figure It Out: Improving the Frontier of Reasoning with Active Visual Thinking | ✓ Link | 79.32 | FigR | 2025-12-30 |
| Introducing OpenAI o1 | | 79.2 | OpenAI o1 | 2024-09-12 |
| OSCAR: Offline Spectral Covariance-Aware Rotation for 2-bit KV Cache Quantization | ✓ Link | 78.89 | OSCAR (GLM-4.7, 358B) | 2026-05-18 |
| Reasoning over mathematical objects: on-policy reward modeling and test time aggregation | | 78.12 | Principia-4B | 2026-03-19 |
| Olmo 3 | ✓ Link | 78.1 | Olmo 3.1 Think 32B | 2025-12-15 |
| Claude Opus 4.1 | | 78.0 | Claude Opus 4.1 | 2025-08-05 |
| Heterogeneous Agent Collaborative Reinforcement Learning | | 75.7 | Qwen3-4B-Instruct (HACPO) | 2026-03-03 |
| Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information | ✓ Link | 75.6 | Qwen3-30B-A3B + AntiSD | 2026-05-12 |
| Hunyuan Dense V1 | ✓ Link | 75.3 | Hunyuan-7B-Instruct | 2025-07-30 |
| Reasoning Cache: Continual Improvement Over Long Horizons via Short-Horizon RL | ✓ Link | 74.9 | RCT-4B + RC | 2026-02-03 |
| Lightning Weave: Improving the Accuracy-Efficiency Frontier of Reasoning Models through Capability Composition | | 74.6 | Lightning Weave (Qwen3.5-4B) | 2026-09-13 |
| Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation | | 74.22 | TGOPD (35B multi-domain mask-only) | 2026-09-02 |
| TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning | | 74.2 | TERMINATOR (Qwen3-14B) | 2026-03-13 |
| D^3-MOPD: Adaptive Dynamic Domain ScheDuling for Efficient Multi-Teacher Distillation | | 74.1 | D3-MOPD Qwen3.6-35B-A3B (benchmark-best checkpoint) | 2026-08-25 |
| Qwen3-Omni Technical Report | ✓ Link | 74.0 | Qwen3-Omni-Flash-Thinking | 2025-09-22 |
| Chain of Mindset: Reasoning with Adaptive Cognitive Modes | ✓ Link | 73.33 | Chain of Mindset (CoM) (Qwen3-VL-32B-Instruct) | 2026-02-10 |
| ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search | ✓ Link | 73.33 | ZGCM-1 | 2026-09-11 |
| Better Supervision Is Nearby: Neighborhood On-Policy Self-Distillation | | 72.78 | N-OPSD (Qwen3-8B) | 2026-09-30 |
| Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning | | 72.22 | OASIS (Qwen3-8B) (extra data) | 2026-09-29 |
| DoubtLess: Training-Free Contrastive Decoding for Efficient Reasoning | ✓ Link | 72.2 | DoubtLess (Qwen3-32B steered by Qwen3-0.6B) | 2026-10-13 |
| Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities | | 72.0 | Gemini 2.5 Flash | 2025-07-07 |
| TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment | | 71.56 | M Blue (14B) | 2026-01-26 |
| TTPO: Test-Time Policy Optimization | ✓ Link | 71.4 | Qwen3-8B + TTPO | 2026-08-27 |
| On-Policy Self-Distillation without Any Supervision | ✓ Link | 71.39 | u-OPSD Qwen3-8B (thinking) | 2026-08-09 |
| SCALE: Selective Resource Allocation for Overcoming Performance Bottlenecks in Mathematical Test-time Scaling | ✓ Link | 71.25 | Qwen3-32B (SCALE) | 2025-11-29 |
| Where the Model Changes Its Mind: Hindsight-Divergence Localization for Efficient Reinforcement Learning with Verifiable Rewards | | 70.56 | HDL (Qwen3-8B) | 2026-09-29 |
| CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning | ✓ Link | 70.42 | CopT (Qwen3-8B) | 2026-05-19 |
| Distilling Long-CoT Reasoning through Collaborative Step-wise Multi-Teacher Decoding | ✓ Link | 70.2 | R1-Qwen (32B) with CoRD (heterogeneous teachers) | 2026-05-04 |
| How to Fine-Tune a Reasoning Model? A Teacher-Student Cooperation Framework to Synthesize Student-Consistent SFT Data | ✓ Link | 70.1 | TESSY | 2026-03-23 |
| ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning | ✓ Link | 69.7 | ThoughtFold (Qwen3-14B) | 2026-06-02 |
| SPADE: Self-Play in Adaptive Synthetic Executable Environments | ✓ Link | 68.8 | SPADE (Games, Qwen3-8B) | 2026-08-19 |
| Rubric-based On-policy Distillation | ✓ Link | 68.75 | ROPD (Qwen3-4B) | 2026-05-08 |
| Not All Tokens Deserve Equal Credit: Counterfactual Sensitivity Credit Reallocation for Long-CoT Reasoning | | 68.3 | CSCR (Qwen3-4B) | 2026-07-30 |
| DataChef: Cooking Up Optimal Data Recipes for LLM Adaptation via Reinforcement Learning | ✓ Link | 66.7 | DataChef-32B | 2026-02-11 |
| OmniOPD: Logit-Free On-Policy Distillation via Speculative Verification | | 65.63 | OmniOPD (Qwen3-4B student, Claude-4.5-haiku teacher) | 2026-05-31 |
| iGRPO: Self-Feedback-Driven LLM Reasoning | | 65.57 | OpenMath-Nemotron-14B + iGRPO | 2026-02-09 |
| Surprising Success, Repeated Failure: Entropy-Guided Credit Assignment for Exploration in LLM Reasoning | ✓ Link | 65.3 | Olmo-3-7B-Think-DPO + EAPO | 2026-09-27 |
| KnowRL: Boosting LLM Reasoning via Reinforcement Learning with Minimal-Sufficient Knowledge Guidance | ✓ Link | 65.21 | KnowRL-Nemotron-1.5B (CSS) | 2026-04-14 |
| SePO: Self-Evolving Prompt Agent for System Prompt Optimization | | 64.22 | SePO-Generalist | 2026-06-03 |
| Efficient Reasoning with Balanced Thinking | ✓ Link | 63.3 | ReBalance (QwQ-32B) | 2026-03-12 |
| Scaling Embeddings Outperforms Scaling Experts in Language Models | | 63.23 | LongCat-Flash-Lite (68.5B) | 2026-01-29 |
| JustRL: Scaling a 1.5B LLM with a Simple RL Recipe | ✓ Link | 62.92 | JustRL-Nemotron 1.5B | 2025-12-18 |
| On-Policy Self-Distillation for Reasoning Compression | ✓ Link | 61.7 | Qwen3-14B (OPSDC) | 2026-03-05 |
| Calibrating Teacher--Student Discrepancy for On-Policy Distillation | | 61.3 | Cal-OPD (Qwen3-30B-A3B-Thinking-2507 -> Qwen3-4B) | 2026-09-18 |
| LongCat-Flash Technical Report | ✓ Link | 61.25 | LongCat-Flash | 2025-09-01 |
| Weak-to-Strong On-Policy Distillation | ✓ Link | 61.0 | W2S-OPD (multi-teacher contrast) | 2026-07-28 |
| Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation | ✓ Link | 60.52 | Waypoint-OPD (Qwen3-14B) | 2026-10-15 |
| ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models | ✓ Link | 60.5 | ThreadWeaver (Qwen3-8B) | 2025-11-24 |
| LLaDA2.0: Scaling Up Diffusion Language Models to 100B | ✓ Link | 60.0 | LLaDA2.0-flash | 2025-12-10 |
| Self-Distillation Zero: Self-Revision Turns Binary Rewards into Dense Supervision | ✓ Link | 60.0 | SD-Zero (Qwen3-4B-Instruct) | 2026-04-13 |
| Which Reasoning Trajectories Teach Students to Reason Better? A Simple Metric of Informative Alignment | ✓ Link | 59.2 | Qwen-3-14B (fine-tuned with Rank-Surprisal Ratio selection) | 2026-01-20 |
| Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation | ✓ Link | 59.2 | Qwen3-8B-Base (Lightning OPD) | 2026-04-14 |
| ExpRL: Exploratory RL for LLM Mid-Training | ✓ Link | 59.07 | ExpRL-Outcome (Ours) | 2026-06-15 |
| SFT Conflicts, RL Coexists: A Theoretical and Empirical Analysis of Multi-Task Learning for LLMs | ✓ Link | 59.0 | Adapted Parallel-RL (DeepSeek-R1-Distill-Qwen-7B) | 2026-08-06 |
| Can We Trust the Teacher? Decoupled Credit Direction-Magnitude for Self-Distillation | | 57.5 | DCSD (Qwen3-4B) | 2026-09-28 |
| Beyond Domain-Routed Supervision: Cross-Teacher Guarding for Multi-Teacher On-Policy Distillation | | 57.4 | Road MOPD (Qwen3-4B) | 2026-10-09 |
| Small RL Controller, Large Language Model: RL-Guided Adaptive Sampling for Test-Time Scaling | | 57.1 | RL-guided (Qwen3-4B-Instruct-Thinking) | 2026-06-02 |
| Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation | ✓ Link | 56.04 | Qwen3-8B + OPRD | 2026-09-08 |
| Filter, Then Reweight: Rethinking Optimization Granularity in On-Policy Distillation | ✓ Link | 55.83 | FiRe-OPD | 2026-06-01 |
| Rethinking Sample Polarity in Reinforcement Learning with Verifiable Rewards | | 54.1 | A3PO (Deepseek-R1-Distill-Qwen-7B) | 2025-12-25 |
| Efficient Reasoning on the Edge | | 54.0 | Qwen2.5-7B-Instruct (OT3, dense) | 2026-03-17 |
| SAF-OPD: Stable Advantage Fusion for On-Policy Distillation | | 53.96 | SAF (Qwen3-4B) | 2026-07-31 |
| Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning | ✓ Link | 53.8 | NPR (Qwen3-4B) | 2025-12-08 |
| MOPD-Router: Rethinking Teacher Routing in Multi-Teacher On-Policy Distillation | ✓ Link | 53.75 | MOPD-Router (ExpertAlign, Qwen3-4B-Non-Thinking student, domain-labeled) | 2026-09-25 |
| Influence-Directed Distillation: Solving the Diversity Bottleneck in Sampled-Token On-Policy Distillation | | 53.3 | Qwen3-4B-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 | 52.92 | RIDE (Qwen3-4B -> Just-Qwen3-4B) | 2026-09-29 |
| Self-Hinting Language Models Enhance Reinforcement Learning | ✓ Link | 52.1 | SAGE (Qwen3-4B-Instruct) | 2026-02-03 |
| LFM2.5-2.6B: Deploy Agents Everywhere | ✓ Link | 51.87 | LFM2.5-2.6B | 2026-08-04 |
| In-Context Reinforcement Learning for Tool Use in Large Language Models | ✓ Link | 51.7 | Qwen3-8B (ICRL) | 2026-03-09 |
| TGRL: Temperature-Grouped Reinforcement Learning for Efficient Exploration in LLMs | ✓ Link | 51.6 | TGRL (Qwen3-32B) (extra data) | 2026-09-27 |
| Nemotron-Labs-Diffusion: A Tri-Mode Language Model Unifying Autoregressive, Diffusion, and Self-Speculation Decoding | | 50.0 | Nemotron-Labs-Diffusion-14B (Diff.) | 2026-07-07 |
| Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models | ✓ Link | 49.8 | Youtu-LLM 2B | 2025-12-31 |
| LSRIF: Logic-Structured Reinforcement Learning for Instruction Following | | 49.6 | Distill-Qwen-14B (LsrIF) | 2026-01-10 |
| Co-Evolving Policy Distillation | | 49.58 | CoPD | 2026-04-29 |
| Kimi K2: Open Agentic Intelligence | ✓ Link | 49.5 | Kimi K2 | 2025-07-28 |
| Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning | | 48.4 | ERC-DAPO (DeepSeek-R1-Distill-Qwen-7B) | 2025-12-05 |
| ARISE: Agent Reasoning with Intrinsic Skill Evolution in Hierarchical Reinforcement Learning | ✓ Link | 48.3 | ARISE + GRPO (Qwen3-4B-Instruct-2507) | 2026-03-17 |
| RISE: Recursive Improvement via Self-Extrapolating Policy Distillation | | 47.9 | RISE (logit,Qwen3-8B,beta0=1.5) | 2026-09-04 |
| Learning from Your Own Mistakes: Constructing Learnable Micro-Reflective Trajectories for Self-Distillation | | 46.88 | Cold-start + TAPO (Qwen3-8B-Instruct) | 2026-06-17 |
| CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning Distillation | | 45.36 | CoTranslator (Qwen3-4B-Thinking-2507) | 2026-10-14 |
| TEMPO: Scaling Test-time Training for Large Reasoning Models | ✓ Link | 44.6 | Qwen3-14B-Base (TEMPO) | 2026-04-21 |
| Trust Region On-Policy Distillation | | 44.06 | TrOPD (Qwen3-SFT-1.7B) | 2026-05-31 |
| Weak-to-Strong Generalization via Direct On-Policy Distillation | ✓ Link | 44.0 | R1-Distill-7B (Direct-OPD) | 2026-07-06 |
| Hybrid Reasoning Agents | ✓ Link | 43.3 | HybridAgent (14B) | 2026-10-13 |
| Every Coin Has Two Sides: On the Dual Nature of Generalization in On-Policy Distillation of Large Language Models | | 43.2 | DS-R1-Distill-Qwen-7B + OPD Dynamic Sampling | 2026-08-17 |
| LFM2.5-8B-A1B: an Even Better on-Device Mixture-of-Experts | ✓ Link | 42.53 | LFM2.5-8B-A1B | 2026-05-28 |
| From P(y|x) to P(y): Investigating Reinforcement Learning in Pre-train Space | ✓ Link | 42.19 | DSRL (Qwen3-8B) | 2026-04-15 |
| Bootstrapping Exploration with Group-Level Natural Language Feedback in Reinforcement Learning | ✓ Link | 41.65 | GOLF (Qwen-3-8B) | 2026-03-04 |
| STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability | ✓ Link | 41.5 | STARE-O1 (DeepSeek-R1-Distill-Qwen-1.5B) | 2026-06-17 |
| MHPO: Modulated Hazard-aware Policy Optimization for Stable Reinforcement Learning | | 41.4 | MHPO (Qwen3-4B-Base) | 2026-03-14 |
| ResRL: Boosting LLM Reasoning via Negative Sample Projection Residual Reinforcement Learning | ✓ Link | 41.1 | ResRL (Qwen3-8B) | 2026-05-01 |
| MiniCPM5-1B | ✓ Link | 40.42 | MiniCPM5-1B (Thinking) | 2026-05-25 |
| TREK: Distill to Explore, Reinforce to Refine | | 40.3 | Qwen3-8B + TREK (DeepSeek-V4 proposals) | 2026-07-06 |
| Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex Reasoning | | 40.0 | Atlas (RL) (pool extension) | 2026-01-07 |
| Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs? | ✓ Link | 40.0 | DeepSeek-R1-Distill-Qwen-7B (SFT on D_ug) | 2026-03-25 |
| GlimpRouter: Efficient Collaborative Inference by Glimpsing One Token of Thoughts | ✓ Link | 39.17 | GlimpRouter (Qwen3-4B SLM + DeepSeek-R1-Distill-Qwen-32B LLM) | 2026-01-08 |
| IIB-LPO: Latent Policy Optimization via Iterative Information Bottleneck | | 38.3 | I2B-LPO (Qwen3-14B) | 2026-01-09 |
| DelTA: Discriminative Token Credit Assignment for Reinforcement Learning from Verifiable Rewards | ✓ Link | 37.92 | DelTA (Qwen3-14B-Base) | 2026-05-20 |
| Recovering Reliable Supervision via Critique-Guided On-Policy Distillation | ✓ Link | 37.81 | CriticOPD (Olmo-3-7B-Instruct-SFT) | 2026-10-14 |
| On the Step Length Confounding in LLM Reasoning Data Selection | ✓ Link | 37.5 | Qwen3-8B-Base (Aslec-casl) | 2026-04-08 |
| Demystifying On-Policy Distillation: Roles, Pathologies, and Regulations | | 37.5 | Qwen3-1.7B (OPD+Clip) | 2026-07-15 |
| Approximation of Log-Partition Function in Policy Mirror Descent Induces Implicit Regularization for LLM Post-Training | ✓ Link | 37.19 | PMD-mean (Qwen3-30B-A3B-Base, τ=0.1) | 2026-02-05 |
| Selection Is Not Learning: Choosing Model Futures | | 37.15 | EvoHeir (Gemma-4-E4B-it) | 2026-10-15 |
| SmolLM3: smol, multilingual, long-context reasoner | ✓ Link | 36.7 | SmolLM3 3B (Extended Thinking) | 2025-07-08 |
| LFM-1B-Math: Can Small Models Be Concise Reasoners? | | 36.25 | LFM-1.3B-Distill (32K output budget) | 2025-06-24 |
| Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoders | ✓ Link | 36.0 | OPD student (JustRL) | 2026-09-28 |
| Prefix Sliding for efficient test-time scaling | ✓ Link | 35.8 | Qwen3-1.7B (Prefix Sliding) | 2026-08-26 |
| PACED: Distillation at the Frontier of Student Competence | | 35.6 | Paced Forward KL (Qwen3-8B) | 2026-03-11 |
| ERNIE 4.5 Technical Report | ✓ Link | 35.1 | ERNIE-4.5-300B-A47B | 2025-06-30 |
| Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability | | 35.1 | Strict top-16 (Qwen3-235B-A22B-Instruct-2507 -> Granite-4.1-8B) | 2026-10-06 |
| LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language Models | ✓ Link | 35.0 | LEAD (DeepSeek-R1-Distill-Qwen-7B) | 2026-05-10 |
| Introducing Claude 4 | | 33.9 | Claude Opus 4 | 2025-05-22 |
| Rethinking On-Policy Distillation of Large Language Models: Phenomenology, Mechanism, and Recipe | ✓ Link | 33.8 | R1-Distill-1.5B + Top-16 OPD | 2026-04-14 |
| MetaAgent-X : Breaking the Ceiling of Automatic Multi-Agent Systems via End-to-End Reinforcement Learning | | 33.33 | MetaAgent-X RL (Qwen3 8B) | 2026-05-14 |
| Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models | | 33.3 | MELT-1.6B | 2026-05-08 |
| Introducing Claude 4 | | 33.1 | Claude Sonnet 4 | 2025-05-22 |
| Composing What Each Teacher Learned: Multi-Teacher On-Policy Distillation through Teacher-Relative Shifts | | 32.92 | Δ-MOPD (1.5B student, Avg@16) | 2026-10-07 |
| TriAttention: Efficient Long Reasoning with Trigonometric KV Compression | ✓ Link | 32.9 | TriAttention (Qwen3-8B) | 2026-04-06 |
| LFM2.5-1.2B-Thinking: On-Device Reasoning Under 1GB | | 31.73 | LFM2.5-1.2B-Thinking | 2026-01-20 |
| Graph-Based Chain-of-Thought Pruning for Reducing Redundant Reflections in Reasoning LLMs | | 31.67 | DeepSeek-R1-Distill-Qwen-7B (Graph-based CoT Pruning) | 2026-04-07 |
| On the Off-Policy Teacher in On-Policy Distillation | | 31.2 | SCOUT | 2026-09-29 |
| Group Adaptive Clipping Policy Optimization | ✓ Link | 30.8 | DeepSeek-R1-Distill-Qwen-1.5B + GAPO | 2026-08-31 |
| APPO: Agentic Procedural Policy Optimization | ✓ Link | 30.0 | APPO (Qwen2.5-7B-Instruct) | 2026-06-10 |
| When Reasoning Meets Its Laws | ✓ Link | 29.2 | DeepSeek-R1-8B (SFT-Compo) | 2025-12-19 |
| On-policy Distillation with Verifiable Reward | ✓ Link | 28.1 | OPDVR (Qwen3-4B) | 2026-08-25 |
| PRISM: Demystifying Retention and Interaction in Mid-Training | | 27.96 | Granite-3.3 (8B) PRISM | 2026-03-17 |
| Towards Full Pipeline FP8 Reinforcement Learning for LLMs | | 27.1 | Qwen3-8B-Base + Calibrated Clipping (Blockwise FP8) | 2026-09-19 |
| Zone of Proximal Policy Optimization: Teacher in Prompts, Not Gradients | | 26.7 | ZPPO (Qwen3.5-2B) | 2026-06-16 |
| Beyond Imitation: Filtering On-Policy Distillation by Reasoning Progress | | 25.83 | R²-OPD (Qwen3-1.7B) | 2026-08-19 |
| Branching On-Policy Distillation | | 25.83 | BranchOPD (extra data) | 2026-10-09 |
| ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy Learning | ✓ Link | 25.2 | ReSPO (Qwen3-30B-A3B-Base, N=8) | 2026-09-28 |
| DenoiseRL: Bootstrapping Reasoning Models to Recover from Noisy Prefixes | ✓ Link | 24.8 | DenoiseRL-DAPO (Qwen3-8B-Base) | 2026-05-27 |
| Qwen3 Technical Report | ✓ Link | 24.7 | Qwen3-235B-A22B (Non-thinking) | 2025-05-14 |
| Understanding Evolution Strategies for LLM Reasoning: Broader Reasoning Coverage than GRPO | | 24.5 | DeepSeek-R1-Distill-Qwen-1.5B (GRPO to ES) | 2026-08-27 |
| Arcee Trinity Large Technical Report | ✓ Link | 24.36 | Trinity Large Preview | 2026-02-19 |
| Normalized Low-Rank Adaptation | ✓ Link | 23.5 | DS-R1-Distill-Qwen1.5B NoRA | 2026-08-31 |
| Your Group-Relative Advantage Is Biased | | 23.3 | DAPO+HA-DW (Qwen-3-8B-Base) | 2026-01-13 |
| BeamPERL: Parameter-Efficient RL with Verifiable Rewards Specializes Compact LLMs for Structured Beam Mechanics Reasoning | ✓ Link | 23.3 | BeamPERL (best-performing checkpoint) | 2026-03-04 |
| TIP: Token Importance in On-Policy Distillation | ✓ Link | 23.2 | TIP Soft-OR 20% (Qwen3-8B → Qwen3-4B) | 2026-04-15 |
| Mitigating the Length-Scaling Tax with Online Distillation | | 22.92 | LSD (SG-FKL) | 2026-09-30 |
| DiRL: An Efficient Post-Training Framework for Diffusion Language Models | ✓ Link | 22.9 | DiRL-8B-Instruct | 2025-12-23 |
| Open-MOPD: Diagnosing and Fixing Capability Imbalance in Multi-Teacher On-Policy Distillation | ✓ Link | 22.86 | Open-MOPD (SmolLM3-3B) | 2026-08-19 |
| Smaller Models are Natural Explorers for Policy-Level Diversity in GRPO | ✓ Link | 22.5 | 1.7B→8B S2L | 2026-06-02 |
| WHALE: A Simple Recipe for Joint Harness-Weight Optimization | ✓ Link | 22.5 | Qwen3.5-2B + WHALE (E0.6,I6) | 2026-08-31 |
| KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization | | 22.1 | KV4-RL (Qwen3-8B-Base) | 2026-10-10 |
| Learning to Solve Hard Problems in RL for LLMs by Never Giving Up | | 22.0 | Qwen 3 4B + NGU (p=0.5) | 2026-09-11 |
| The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents | | 21.1 | Qwen2.5-3B with CAR (MSCR) | 2026-01-12 |
| Can LLMs Learn to Reason Robustly under Noisy Supervision? | ✓ Link | 20.8 | Qwen3-4B-Base (with OLR) | 2026-04-05 |
| Can LLMs Guide Their Own Exploration? Gradient-Guided Reinforcement Learning for LLM Reasoning | | 20.1 | G2RL (Qwen3-4B-Base) | 2025-12-17 |
| SPARK: Stepwise Process-Aware Rewards for Reference-Free Reinforcement Learning | | 20.0 | PRM-CoT (Process-Aware) (Qwen2.5-Math-7B) | 2025-12-02 |
| nabla-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space | ✓ Link | 20.0 | ∇-Reasoner (Qwen-2.5-7B-Instruct) | 2026-03-05 |
| Sparse but Critical: A Token-Level Analysis of Distributional Shifts in RLVR Fine-Tuning of LLMs | | 20.0 | Qwen2.5-Math-7B with DAPO (High-KL boost) | 2026-03-23 |
| Unlocking Token Rewards via Training-Free Reward Attribution | | 20.0 | Qwen2.5-Math-7B + GRPO (P2T token reward) (extra data) | 2026-06-03 |
| LLaDA MoE v2: Scaling Mixture-of-Experts Diffusion Language Models | | 20.00 | LLaDA MoE v2 Instruct | 2026-08-04 |
| Multiplex Thinking: Reasoning via Token-wise Branch-and-Merge | ✓ Link | 19.7 | Multiplex Thinking (DeepSeek-R1-Distill-Qwen-7B) | 2026-01-13 |
| Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation | ✓ Link | 19.4 | OPDLM-8B (think@eval) | 2026-06-04 |
| SPOT: Sparse Probing and Outcome Calibration for On-Policy Distillation | ✓ Link | 17.92 | SPOT Qwen3-4B-Base | 2026-08-05 |
| Think Anywhere in Code Generation | ✓ Link | 17.7 | Think-Anywhere | 2026-03-31 |
| Adaptive Ability Decomposing for Unlocking Large Reasoning Model Effective Reinforcement Learning | | 17.1 | A2D (Qwen2.5-7B-Instruct) | 2026-01-31 |
| Low-rank Optimization Trajectories Modeling for LLM RLVR Acceleration | ✓ Link | 16.7 | Qwen2.5-14B-Instruct with NExt | 2026-04-13 |
| Falcon-H1: A Family of Hybrid-Head Language Models Redefining Efficiency and Performance | ✓ Link | 16.67 | Falcon-H1-34B-Instruct | 2025-07-30 |
| J-Zero: Unified Challenger--Solver--Judge Co-Evolution from Zero Data | ✓ Link | 15.94 | J-Zero-8B | 2026-08-27 |
| Learning to Hint for Reinforcement Learning | ✓ Link | 15.3 | HiLL (Qwen2.5-7B-Instruct) | 2026-04-01 |
| From Reasoning Chains to Verifiable Subproblems: Curriculum Reinforcement Learning Enables Credit Assignment for LLM Reasoning | | 15.3 | SCRL (Qwen3-4B-Base) | 2026-05-21 |
| Long Live The Balance: Information Bottleneck Driven Tree-based Policy Optimization | ✓ Link | 15.3 | IBTPO (Qwen3-8B-Base) | 2026-05-27 |
| Boosting Reinforcement Learning with Verifiable Rewards via Randomly Selected Few-Shot Guidance | ✓ Link | 14.58 | FEST-GRPO | 2026-05-14 |
| Dynamic Important Example Mining for Reinforcement Finetuning | ✓ Link | 14.2 | Qwen3-4B + DIEM | 2026-08-29 |
| Introducing LFM2.5: The Next Generation of On-Device AI | | 14.00 | LFM2.5-1.2B-Instruct | 2026-01-05 |
| Stratagem: Learning Transferable Reasoning via Trajectory-Modulated Game Self-Play | ✓ Link | 13.3 | Stratagem (Kuhn Poker) | 2026-04-20 |
| Solving Physics Olympiad via Reinforcement Learning on Physics Simulators | ✓ Link | 12.5 | Qwen2.5-32B + RL (synthetic) | 2026-04-13 |
| G-Zero: Self-Play for Open-Ended Generation from Zero Data | ✓ Link | 12.4 | Qwen3-8B-Base (G-Zero R2) | 2026-05-11 |
| Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning | | 11.7 | Qwen2.5-Math-7B + ACE-DAPO | 2026-02-24 |
| Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVR | ✓ Link | 10.9 | DATPO (Qwen3-4B-Base) | 2026-09-08 |
| Enabling Token-wise Importance Sampling in Off-Policy RLVR with Proximal KL Regularization | ✓ Link | 10.63 | Proximal KL (Qwen3-4B-Base, off-policy, staleness 4, lr 5e-6) | 2026-10-13 |
| Gains and Collapse in On-Policy Distillation:A Reinforcement Learning Perspective | ✓ Link | 10.0 | 4B → SFT (warmup) | 2026-10-02 |
| The Molecular Structure of Thought: Mapping the Topology of Long Chain-of-Thought Reasoning | | 8.33 | Llama-3.1-8B-Instruct + Mole-Syn-by-Qwen-32B | 2026-01-09 |
| Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models | | 7.4000 | WeDLM-8B (IMH) | 2026-04-01 |
| Trust-Region Behavior Blending for On-Policy Distillation | | 6.9 | Trust-Region behavior Blending (Qwen3-1.7B-Base ← Qwen3-8B) | 2026-05-29 |
| LastOPD: Taming Collapse in Latent On-Policy Distillation | ✓ Link | 4.58 | LastOPD (Qwen3-8B -> Qwen3-1.7B-Base) | 2026-09-23 |
| SLPO: Scaling Latent Reasoning via a Surrogate Policy | ✓ Link | 3.33 | SLPO (Llama-3.2-3B-Instruct) | 2026-07-22 |
| Mid-Training with Self-Generated Data Improves Reinforcement Learning in Language Models | | 0.44 | Llama 3.2–3B–Instruct (mid-trained, n=64) | 2026-05-08 |