116 lines (89 loc) · 8.94 KB

Awesome-RL-Reasoning

Papers

Algorithm

(2026.02) Maximum Likelihood Reinforcement Learning, [Tajwar+]
(2025.12) Stabilizing Reinforcement Learning with LLMs: Formulation and Practices, [Zheng+]
(2025.10) Stabilizing MoE Reinforcement Learning by Aligning Training and Inference Routers, [Ma+]
(2025.08) Sample More to Think Less: Group Filtered Policy Optimization for Concise Reasoning, [Shrivastava+]
(2025.08) Part I: Tricks or Traps? A Deep Dive into RL for LLM Reasoning, [Liu+, ICLR 2026]
(2025.07) Group Sequence Policy Optimization, [Zheng+]
(2025.06) Truncated Proximal Policy Optimization, [Fan+]
(2025.04) VAPO: Efficient and Reliable Reinforcement Learning for Advanced Reasoning Tasks, [Yue+]
(2025.03) Understanding R1-Zero-Like Training: A Critical Perspective, [Liu+, COLM 2025]
(2025.03) DAPO: An Open-Source LLM Reinforcement Learning System at Scale, [Yu+, NeurIPS 2025]
(2024.02) DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models, [Shao+]

Scaling

(2025.10) The Art of Scaling Reinforcement Learning Compute for LLMs, [Khatri+]
(2025.06) AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergy, [Liu+]
(2025.05) ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models, [Liu+, NeurIPS 2025]
(2025.05) AceReason-Nemotron: Advancing Math and Code Reasoning through Reinforcement Learning, [Chen+, NeurIPS 2025]

Asynchronous

(2025.09) PipelineRL: Faster On-policy Reinforcement Learning for Long Sequence Generation, [Piché+]
(2025.05) AReaL: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning, [Fu+, NeurIPS 2025]

Technical Report

(2026.04) JoyAI-LLM Flash: Advancing Mid-Scale LLMs with Token Efficiency, [JD.com]
(2026.03) Composer 2 Technical Report, [Cursor Research]
(2026.02) GLM-5: from Vibe Coding to Agentic Engineering, [GLM-5-Team]
(2026.02) Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts, [Yang+]
(2026.02) Kimi K2.5: Visual Agentic Intelligence, [Kimi Team]
(2025.12) MiMo-V2-Flash Technical Report, [LLM-Core Xiaomi]
(2025.11) Olmo 3, [Olmo Team]
(2025.10) Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model, [Ling Team]
(2025.09) DeepSeek-V3.2-Exp: Boosting Long-Context Efficiency with DeepSeek Sparse Attention, [DeepSeek-AI]
(2025.09) CWM: An Open-Weights LLM for Research on Code Generation with World Models, [Meta FAIR CodeGen Team]
(2025.09) LongCat-Flash-Thinking Technical Report, [Meituan LongCat Team]
(2025.08) GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models, [GLM-4.5 Team]
(2025.07) Kimi K2: Open Agentic Intelligence, [Kimi Team]
(2025.06) Hunyuan-A13B Technical Report, [Tencent Hunyuan Team]
(2025.06) MiniMax-M1: Scaling Test-Time Compute Efficiently with Lightning Attention, [MiniMax]
(2025.06) Magistral, [Mistral-AI]
(2025.05) Skywork Open Reasoner 1 Technical Report, [Skywork AI]
(2025.05) Qwen3 Technical Report, [Qwen Team]
(2025.05) Llama-Nemotron: Efficient Reasoning Models, [NVIDIA]
(2025.04) Phi-4-reasoning Technical Report, [Microsoft]
(2025.04) Seed1.5-Thinking: Advancing Superb Reasoning Models with Reinforcement Learning, [ByteDance Seed]
(2025.01) DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning, [DeepSeek-AI]
(2025.01) Kimi k1.5: Scaling Reinforcement Learning with LLMs, [Kimi Team]

Blogs

Deterministic and Reproducibility

(2025.11) No More Train-Inference Mismatch: Bitwise Consistent On-Policy Reinforcement Learning with vLLM and TorchTitan, [vLLM and TorchTitan Teams]
(2025.09) Towards Deterministic Inference in SGLang and Reproducible RL Training, [The SGLang Team]
(2025.09) Defeating Nondeterminism in LLM Inference, [Horace He and Thinking Machines Lab]

Training–Inference Mismatch

(2025.09) When Speed Kills Stability: Demystifying RL Collapse from the Training-Inference Mismatch, [Liu+]
(2025.08) Your Efficient RL Framework Secretly Brings You Off-Policy RL Training, [Yao+]

Scaling and Open-Source

(2025.08) ProRL V2 - Prolonged Training Validates RL Scaling Laws, [Hu+]
(2025.04) DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level, [Agentica x Together AI]
(2025.02) DeepScaleR: Surpassing O1-Preview with a 1.5B Model by Scaling RL, [Luo+]

Frameworks