rl — AI Digest

17 июн VibeThinker-3B Reaches Frontier-Level Reasoning Benchmarks via Curriculum RL WeiboAI research

18 июл OpenAI unveils GPT-Red, an internal automated red-teaming model for prompt-injection defense OpenAI research

19 авг StateM: harness scaling pushes GPT-5.6 Sol to 95.3% on Terminal-Bench 2.1 and the same harness adapts to DeepSeek-V4 Flash for $15 total spend research

3 мая Exploration Hacking: LLMs Can Be Fine-Tuned to Strategically Resist RL Training research

9 мая OpenAI Discloses Accidental Chain-of-Thought Grading in RL Training Across Six Models OpenAI research

10 мая Google DeepMind's AI Co-Mathematician Reaches 48% on FrontierMath Tier 4 Google DeepMind research

11 мая Flow-OPD: On-Policy Distillation Pushes GenEval +29 Points on Stable Diffusion 3.5 research

13 мая RubricEM: Meta-RL with Rubric-Guided Policy Decomposition Beyond Verifiable Rewards Google research

15 мая SU-01: Gold-Medal-Level Olympiad Reasoning via Curriculum SFT and Two-Stage RL SU-01 Team research

20 мая SkillsVote: Lifecycle Governance of Agent Skills — Collection, Recommendation, Evolution (219 HF upvotes) Memtensor Research Group / IAAR-Shanghai research

26 июн Qwen-AgentWorld: Language World Models for General Agents at 35B and 397B Scale Qwen Team, Alibaba research

2 июл Dockerless: Environment-Free Program Verifier for Coding Agents ByteDance research

2 июл DOPD: Dual On-Policy Distillation with Advantage-Aware Token Routing research

7 июл The Mirage of Optimizing Training Policies: Monotonic Inference Policies as the Real Objective for LLM Reinforcement Learning Tianjin University / Alibaba research

13 июл Long-Horizon-Terminal-Bench: Testing Agent Limits on Long-Horizon Terminal Tasks Tencent Hunyuan research

27 июл Molt: A Scalable PyTorch-Native Training Framework for Agentic Reinforcement Learning NVIDIA research

16 авг Intern-S2-Preview: 397B scientific agentic foundation model from Shanghai AI Lab Shanghai AI Lab research

18 авг ClawGym II: Exploring Black-Box RL on Agent Harness Renmin University of China research

28 авг WarpSAC: Towards the Pinnacle of Scalable Off-policy RL by Rethinking Exploration and Exploitation research

14 сен Recurrent Looped Transformer: viral technical report claims 'infinite temporal depth' for latent reasoning Princeton University research

15 сен DataFlex-RL: evaluation platform finds RLVR data policies don't beat uniform sampling Peking University research

22 сен CodeMidas: Scaling Agentic Coding RL Environments from Code Itself Xiaomi MiMo research

10 мая Anthropic Eliminates Claude's Agentic Blackmail Behavior via 'Teaching Claude Why' Anthropic research

10 июн DRPO: Rethinking Divergence Regularization in LLM Reinforcement Learning Tencent Hunyuan research

8 июл LLM-as-a-Verifier: verification as an independent scaling axis for LLMs Stanford University / UC Berkeley / NVIDIA research

29 июл Kimi K3 Technical Report: Kimi Delta Attention and Stable LatentMoE Architecture Detailed Moonshot AI research

29 авг Agentic Game Development as a Verifiable Trajectory Data Engine for Scaling World Models NUS research

5 сен Terminal-Universe: turning agent trajectories into scalable terminal environments research

12 сен WMRL: scaling automatic research agents via world models University of Illinois / NVIDIA research

13 сен NVIDIA Open Recipe for IMO Gold with Natural-Language Nemotron Pipeline NVIDIA research

16 сен Dream-RSI: recursive self-improvement through evolving worlds Google research

23 окт Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds Tsinghua THUSI-Lab, miHoYo research

4 мая Learning while Deploying: Fleet-Scale Reinforcement Learning Turns Robot Deployment into Continuous Training AGIBot research

6 мая Ctx2Skill: Self-Improving Framework for Autonomous Context-Skill Discovery in LLMs research

7 мая RLDX-1: Multi-Stream Action Transformer Achieves 86.8% on ALLEX Humanoid Tasks RLWRLD research

9 мая OpenSearch-VL: Open Recipe for Training Frontier Multimodal Search Agents Tencent Hunyuan research

16 мая SDAR: Self-Distilled Agentic Reinforcement Learning for Multi-Turn Agents Zhejiang University / Meituan research

2 июн GrepSeek: Training Search Agents for Direct Corpus Interaction via Shell Commands (93 HF Upvotes) University of Massachusetts Amherst research

4 июн ThoughtFold: Introspective Preference Learning Cuts Reasoning Tokens by 56% Without Accuracy Loss research

8 июн Agentic Transformers Provably Learn Depth-First Search via Reinforcement Learning Carnegie Mellon University / Ohio State University research

10 июн Flow-DPPO: Principled RL Alignment for Flow Matching Image and Video Models Tencent Hunyuan research

11 июн Arbor: Generalist Autonomous ML Research via Hypothesis-Tree Refinement NLPIR Lab research

11 июн Z-Reward: Score Distributions Instead of Scalar Rewards for Image Generation RLHF Alibaba research

14 июн InterleaveThinker: RL Planner+Critic Pipeline for Interleaved Text-and-Image Generation CUHK Multimedia Lab research

29 июн OPID: On-Policy Skill Distillation Improves Long-Horizon Agent RL Institute of Automation, Chinese Academy of Sciences research

7 июл CompactionRL: Reinforcement Learning with Context Compaction for Long-Horizon Agents Zhipu AI / Tsinghua University research

17 июл SearchOS-V1: Towards Robust Open-Domain Information-Seeking Agent Collaboration Ant Group / Renmin University research

17 июл SEED: Self-Evolving On-Policy Distillation for Agentic Reinforcement Learning research

19 июл From Pixels to States: Rethinking Interactive World Models as Game Engines research

23 июл Beyond Euclidean Clipping: Overcoming Exploration Collapse in LLM RL via Riemannian Isometric Policy Optimization research

2 авг Frontis-MA1: Training an AI4AI Model towards Recursive Self-Improvement in Machine Learning Engineering Horizon Research, Frontis.AI, Tsinghua University research

5 авг DAPD: Dual-Anchored Policy Distillation research

5 авг Progressive Agent Skill Generation via Reinforcement Learning (Skill-α) The Chinese University of Hong Kong research

9 авг Recursive Synthetic Terminal Tasks generates 37,000+ verifiable long-horizon agent tasks research

9 авг DAPD identifies 'privilege illusion' failure mode in on-policy self-distillation Shanghai AI Laboratory research

10 авг AgentOPSD: Recursive Self-Distillation for Agentic Reinforcement Learning Tsinghua University research

13 авг On-Policy Self-Distillation without Any Supervision research

23 авг EnvHarness: Awakening Static Worlds for Agent Learning Google research

23 авг FACET: Preserving Source Intent and Executable State in Terminal Task Synthesis University of Science and Technology of China research

29 авг TTPO: Test-Time Policy Optimization Zhejiang University research

29 авг Self-OPD: On-Policy Distillation for Flow Matching Models without Teacher research

8 сен Iris: Climbing to the Search Frontier research

10 сен Miles v0.1: a production-level RL post-training system RadixArk research

11 сен Cognition announces SWE-2 coding model, near-parity with Fable 5.1 at 64% lower cost Cognition models-llm

18 сен Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening Shanghai AI Laboratory research

23 сен Intern-S2-397B: Shanghai AI Lab's strongest open scientific multimodal model Shanghai AI Laboratory (InternLM) models-llm

23 сен RRSI: Regularized Recursive Self-Improvement of Agent Harnesses Google Research research

17 окт Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation research

24 окт IQuest-Q1: Advancing Foundation Capabilities for Agentic CLI Systems IQuest models-llm

2 мая CoPD: co-evolving policy distillation for unified multi-capability models research

5 мая Odysseus: Training VLMs for 100+ Turn Interactive Decision-Making via RL Princeton University research

3 июн TrOPD: Trust-Region On-Policy Distillation Stabilizes LLM Training When Teacher-Student Gap Is Large Samsung Research research

12 июн InterleaveThinker: RL Framework for Agentic Text-and-Image Interleaved Generation research

12 июн FORT-Searcher: Shortcut-Resistant Training Data Framework for Deep Search Agents research

12 июн Astra: RL-Trained VLM Queries World Simulator for Spatial Reasoning research

28 июн The Verification Horizon: No Single Reward Function Works for Coding Agents at Scale Qwen (Alibaba) research

6 июл EvoPolicyGym: Evaluating Iterative RL Policy Self-Improvement by Coding Agents University of Macau / CUHK research

30 июл Pass the Baton: Trajectory-Relayed On-Policy Distillation Zhejiang University research

17 авг DarwinX: Evolving Agent Harnesses Through Natural Selection Salesforce AI Research research

18 авг VibeWorlding: Can Multimodal Agents Construct 3D Open Worlds End-to-End? Tencent research

25 авг Prime Agent: A Self-Improving RLM Harness Prime Intellect research

22 сен RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents Qwen (Alibaba) research

6 мая HeavySkill: Internalizing Heavy Thinking as a Trainable Agentic Skill via RL research

12 мая NanoResearch: Co-Evolving Skills, Memory, and Policy for Personalized AI Research Automation Shanghai AI Lab research

12 мая TMAS: Scaling Test-Time Compute via Multi-Agent Synergy with Hierarchical Memory research

18 мая BetaPRM: Uncertainty-Aware Process Rewards Cut Reasoning Token Use by 33% research

19 мая NudgeRL: Strategy-Level Context Nudges for Efficient RLVR Exploration KAIST AI research

3 июн QUBRIC: Co-Designing Queries and Rubrics Extends RLVR to Open-Ended Reasoning Domains research

9 июн On the Geometry of On-Policy Distillation: A Training Paradigm Distinct from SFT and RLVR Hong Kong University of Science and Technology research

9 июн Weak Critics Make Strong Learners: On-Policy Critique Distillation for Scalable Oversight Rutgers University research

17 июн ZPPO: Teacher-in-Prompts Knowledge Distillation Outperforms Gradient Methods for Small Reasoners NVIDIA research

23 июн Agentic Transformers Provably Learn to Search via Reinforcement Learning research

26 июн OPRD: On-Policy Representation Distillation for Post-Training LLMs research

4 июл VRRL: Visually Grounded Self-Reflection for Vision-Language Models via RL UT Austin / Cornell research

7 июл Weak-to-Strong Generalization via Direct On-Policy Distillation ByteDance / Tsinghua University research

9 июл RL Post-Training Actively Builds Compositional Reasoning Strategies, Not Just Amplifies Base Skills research

18 июл LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget Mind Lab research

18 июл Ring-Zero: Scaling Zero RL to a Trillion Parameters for Emergent Reasoning inclusionAI research

21 июл Understanding Reasoning from Pretraining to Post-Training research

24 июл AREX: Towards a Recursively Self-Improving Agent for Deep Research Beijing Academy of Artificial Intelligence (BAAI) research

29 июл Skill Self-Play: Pushing the Frontier of LLM Capability with Co-Evolving Skills Alibaba research

3 авг From RLVR to RLSVR: Task Transformation Induces Self-Verifiable Rewards for Open-Ended LLM Self-Improvement research

4 авг Weak-to-Strong On-Policy Distillation Microsoft Research / University of Maryland / MBZUAI research

7 авг MerchantBench: Benchmarking LLM Agents for Long-Term Coherence in E-Commerce Operations Alibaba Group research

7 авг ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment Shanghai Jiao Tong University research

20 авг Agentic ESOpt: Fine-Tuning Long-Horizon LLM Agents with Minimal GPU Requirements National University of Singapore (Zhi Zheng, Wee Sun Lee et al.) research

20 авг Co-RL: Unsupervised Reasoning Emerges from Diverse Cohort in Multi-agent RL UC San Diego (Yunhao Yang, Nuno Vasconcelos, Yijiang Li et al.) research

29 авг Understanding Evolution Strategies for LLM Reasoning: Broader Reasoning Coverage than GRPO research

1 сен DreamX-Creator: native joint audio-video generation at 2K resolution AMAP-ML (Alibaba) research

1 сен Does on-policy distillation really distill? Teacher-free OPSA beats it on AIME24 Purdue University research

1 сен Survey: scaling reasoning models beyond human supervision via a five-level autonomy ladder research

9 сен FlowBalance: verifier-grounded self-improvement for reasoning models research

10 сен OPRD: eliciting weak-to-strong generalization with on-policy reverse distillation KAIST AI research

25 сен SpeakerMem-R1: speaker-centered dual-track memory for multi-party dialogue Zhejiang University research

28 сен Raschka: focus on post-training open-weight LLMs, not pre-training research

28 апр Agentic World Modeling: Foundations, Capabilities, Laws, and Beyond HKUST/NUS/Oxford/NTU research

28 апр World-R1: Reinforcing 3D Constraints for Text-to-Video Generation Microsoft Research research