https://daily.jiayutool.cn/papers/ 每日筛选 agent、模型、推理与对齐方向论文，基于 arXiv 全文深读。 zh-cn Mon, 12 Oct 2026 08:09:48 +0800 https://daily.jiayutool.cn/papers/2026-10-12/ https://daily.jiayutool.cn/papers/2026-10-12/ Mon, 12 Oct 2026 08:09:48 +0800 Learning to Steer, Steering to See: Unveiling the Geometry of RLVR in Large Language Models via Trainable Vectors；BrickBench: Evaluating Agentic Brick Design；Pumpire: Unified Benchmark for Metric Distance Estimation；Can AI Agents Make Open-Ended Scientific Discovery? Evidence from Station；SpaceFlow: Locally Controllable 3D Generation；Frozen Models, Evolving Expertise: Model-Agnostic Learning from Deployment Experience for Multimodal Medical AI；Post-Training Frontier Text-to-Image Models by Composing Preference and Rubric Rewards；Distilling Routed 3D Privilege for Spatial Reasoning in Vision-Language Models https://daily.jiayutool.cn/papers/2026-10-11/ https://daily.jiayutool.cn/papers/2026-10-11/ Sun, 11 Oct 2026 08:07:51 +0800 A GPU-Parallel Framework for Heterogeneous Multi-Task Reinforcement Learning；MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement；Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching；Opera: A Verbal Critic Framework for Long-horizon Coding Agents；TestPrism: Rethinking Test Evaluation Beyond a Single Reference；Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks；WorldGuide: Goal-Directed Video World Model for Procedural Task Execution；REMORY: Learning Residual Memory for Context Compaction https://daily.jiayutool.cn/papers/2026-10-10/ https://daily.jiayutool.cn/papers/2026-10-10/ Sat, 10 Oct 2026 08:11:16 +0800 Foundations of Large Language Models；Multi-Agent Egocentric World Model with Fine-Grained Embodied Interaction；MC-Sparse: Deconstructing and Closing the Dense-Sparse Attention Gap in Diffusion Transformers；OuroWorld: Bringing Any 3D World Alive as Diverse, Endlessly Looping 3D Cinemagraphs；Mara Chain: Rethinking Failure as a Stepping Stone for AI System Auto-Evolution；U-Space: Uncovering When and Why Uncertainty Arises in Language Models；OneSearch-VL: Unified Multimodal Deep Research Agent for Image and Video；SparseEngine: Sparse-First Inference Engine https://daily.jiayutool.cn/papers/2026-10-09/ https://daily.jiayutool.cn/papers/2026-10-09/ Fri, 09 Oct 2026 08:06:27 +0800 STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantization；UniWAM: Unified World-Action Model；GRACE: Generation-aware latent compression for efficient video generation；UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation；Tetris3D: 3D Scene Generation With Objects That Fit Together；RobotWorld: Benchmarking Multimodal Agents for Robot Use Across Diverse Tasks and Embodiments；VepAgent: Bridging Causal-Transition via Tool-Augmented Reinforcement Learning for Video Event Prediction；MIMESIS: Learning User Simulators as Training Environments for Interactive Agents https://daily.jiayutool.cn/papers/2026-10-08/ https://daily.jiayutool.cn/papers/2026-10-08/ Thu, 08 Oct 2026 08:10:06 +0800 NeMo-DCR: Bit-Exact Delta-Compressed Refit for Scalable Agentic RL at Trillion-Parameter Scale；CheckerBench: Can Long-Horizon Agents Synthesize Static-Analysis Checkers?；From Evidence to Action: How Tool-Using Agents Fail；TRACE: Rollout-Guided Quantization-Aware Training for FP4 Reinforcement Learning of MoE Language Models；Rethinking Cross-Tokenizer On-Policy Distillation: From Alignment Coverage to Supervision Reliability；DMAD: Distribution Matching as Adversarial Distillation for Fast Visual Generation；DAEDALUS: Bootstrapping Agent Memory from Self-Generated Tasks；AdvSim2Real : Training Web Agents Against Adaptive Prompt Injection in a Web World Model https://daily.jiayutool.cn/papers/2026-10-07/ https://daily.jiayutool.cn/papers/2026-10-07/ Wed, 07 Oct 2026 08:08:10 +0800 LMBuild: Evaluating LLM Agents for Generating Buildable and Functional Structures；In-Distribution Forcing for Long Video Generation at Test Time；Representation-Space MMD for Diffusion Language Models；RobotUse: Allocating Computation, Context, and Decisions；Optimizing the Optimizer: Language Models Discover Faster Molecular Relaxation；OmniReasoning: Pushing the Limits of Audio-Visual Joint Reasoning；SearchJev: A Fast and Calibrated System-1 Model for Search Agents；Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation https://daily.jiayutool.cn/papers/2026-10-06/ https://daily.jiayutool.cn/papers/2026-10-06/ Tue, 06 Oct 2026 08:14:17 +0800 RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations；VeriHarness: Scaling Agentic Verification for Long-Horizon Tasks；HelixWorld: A Real-time Interactive Audio-Visual World Model；Does Learning Protein Folding Generalize to Broader Reasoning?；SimuVerity: Benchmarking Agents for Engineering-Grade Simulink Model Generation；Science Utopia? Closed-Loop LLM Simulation of Academic Research Ecosystems；Science or Slop?: Benchmarking and Mitigating Scientific Slop in AI-Generated Papers；4DCodeBench: Benchmarking Agents on Inverse Graphics of Dynamic Scenes https://daily.jiayutool.cn/papers/2026-10-05/ https://daily.jiayutool.cn/papers/2026-10-05/ Mon, 05 Oct 2026 08:10:35 +0800 Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs；CorrGRPO: Correlation-Normalized GRPO for Multi-Reward Learning；Decoding Looped Transformers Better for (Almost) Free；SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation；Persona Dosing: Calibrated Activation Steering for Graded Trait Control；Omni-Embed-Mini: Binding Modalities Without Forgetting via Dense Distillation；PhysVista: Benchmarking Physical Intelligence in VLMs via a Perception-Reasoning-Assessment Loop；Does Native 3D Texture Generation Necessarily Require 3D Assets for Training? https://daily.jiayutool.cn/papers/2026-10-04/ https://daily.jiayutool.cn/papers/2026-10-04/ Sun, 04 Oct 2026 08:09:17 +0800 X-Tree: Tokenizing Reusable Experience for Efficient Agent Generalization；Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It；RLE-Bench: A Qualifying Exam for Coding Agents as Robot Learning Engineers；Multimodal Flow: Unified Flow Modeling of Language and Vision in Embedding Spaces；Argo-Bench: Evaluating Data Agents on Enterprise-Scale Workflows；Ego2Act: Evaluating Goal-Directed Manipulation in Egocentric Video Generation；InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation；KaliBench: A Fine-Grained Benchmark for Cybersecurity Tool Use on Kali Linux with Runtime-Free Verifiable Rewards https://daily.jiayutool.cn/papers/2026-10-03/ https://daily.jiayutool.cn/papers/2026-10-03/ Sat, 03 Oct 2026 08:06:06 +0800 OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video Interaction；On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics；Fewer Tokens, Better Action: GPT-6 Astra Robot Agents with 14% Higher Success Rate but 65% Fewer Tokens；Video Generation Models: A Survey of Post-Training and Alignment；EgoTools: Towards Tool-Centric Reasoning in Real-World Egocentric Videos；Decentralized Master-Mind: Joint Action Refinement through Iterative Intent Denoising in Multi-Agent Pathfinding；Make Sparse Rewards Count: Density-Aware Reward Aggregation for Multi-Reward RL；DataMagic: Authoring Data Videos through Declarative Multi-Agent Orchestration https://daily.jiayutool.cn/papers/2026-10-02/ https://daily.jiayutool.cn/papers/2026-10-02/ Fri, 02 Oct 2026 08:05:31 +0800 Rethinking Latent Visual Reasoning: Grounding Latent Reasoning in Visual Evidence；False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents；EVOKE: Eliciting World Knowledge in Agents for Transferable Decision-Making；Mid-Harness: Scaling Actions Between Model and Harness for Terminal Agents；OSWorld-Science: A Benchmark of Computer Use Agents for Learning and Using Scientific Software；TERRA: Terrain-Aware Reconstruction, Retargeting and Control for Musculoskeletal Locomotion；Breaking Babel: A Self-Evolving Multi-Agent System for Long-Form Subtitle Translation；MILO: Automated Harness Discovery via Orchestrated Multi-Agent Evolution https://daily.jiayutool.cn/papers/2026-10-01/ https://daily.jiayutool.cn/papers/2026-10-01/ Thu, 01 Oct 2026 08:10:18 +0800 Omni-IO Skills: Harnessing Your Agent Omni-Native；EngiWorld: What Can Frontier Agents Deliver in Professional Engineering Environments?；LLMs are General Asynchronous Agents；LongCat-DeepResearch Technical Report；Scaling Properties of Same-Family On-Policy Distillation；VoxMem: Benchmarking Multimodal Memory in Large Audio Language Models；Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression；Context Language Models https://daily.jiayutool.cn/papers/2026-09-30/ https://daily.jiayutool.cn/papers/2026-09-30/ Wed, 30 Sep 2026 08:19:41 +0800 VisionHOPE: Visual Backbones as Self-Modifying Learning Systems；Learning Native Reflection in Unified Models with Interleaved Reinforcement Learning；Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL；How Far Are We from Removing the Visual Encoder? Scaling Laws for Encoder-Free Multimodal Pretraining；QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents；WideSWE: Can Coding Agents Coordinate Changes Across Repositories?；TT-VidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining；Recursive Harness Distillation across Agents for Robot Manipulation https://daily.jiayutool.cn/papers/2026-09-29/ https://daily.jiayutool.cn/papers/2026-09-29/ Tue, 29 Sep 2026 08:05:28 +0800 FuseReg: Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders；Disaggregated Quantization: Specializing LLM Prefill and Decode；RayOrch: Programming and Executing Lineage-Controlled Multi-Grain Dataflows for Foundation-Model Data Preparation；SLCA-GRPO: Resolving Cross-Segment Credit Misattribution in Tool-Calling RL；Block Sparse Attention with Log-Linear Complexity；VLA-Precision: Asymmetric Co-Bootstrapping for Efficient Real-World Online RL of Vision-Language-Action Models；Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors；FoMo: Forking Moment in Generative Trajectory as a Perceptual Distance https://daily.jiayutool.cn/papers/2026-09-28/ https://daily.jiayutool.cn/papers/2026-09-28/ Mon, 28 Sep 2026 08:20:10 +0800 AV-GRPO: Modality-Anchored Decoupling Diffusion Reinforcement Learning for Joint Audio-Video Generation；Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures；X-Planner: Event-Structured Task Planning for Embodied Intelligence；WanPE: Towards Cinematic Prompt Enhancement for Modern Text-to-Video Generation；Hunyuan-A13B Technical Report；Learning to Discover Interesting Mathematics；Qwen-Planner-Agent: A Closed-Loop AI-for-AI Framework for Real-World Mobile Planner Agents；PACT: From Credit Assignment to Critic Alignment https://daily.jiayutool.cn/papers/2026-09-27/ https://daily.jiayutool.cn/papers/2026-09-27/ Sun, 27 Sep 2026 08:18:45 +0800 Rufus-Air: An Open LLM Post-Training Recipe；RGBD20K: A Large-Scale Benchmark for RGB-D Semantic Segmentation；WhatWorkedBench: Benchmarking Experimental Understanding in AI Agents；Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanisms；Neural Spectral Capacity: Measuring and Designing Architectures from Network Specification Alone；EmbodiedSWE: Coding Agents for Long Horizon Dexterous Robotics；Parts-of-Speech as Emergent Categories in SAE Latent Space；ExplorationBench: Measuring AI Systems' Exploration in Verifiable Alien Worlds https://daily.jiayutool.cn/papers/2026-09-26/ https://daily.jiayutool.cn/papers/2026-09-26/ Sat, 26 Sep 2026 08:08:17 +0800 Training Object Permanence in World Models；OmniEcho: Spatial Audio Understanding for Embodied Agents；Coding Agents for Generalized Task and Motion Planning Problems；Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs；PackLab: A Comprehensive Framework for Developing, Training, and Evaluating MLLMs in Robotic Bin Packing；Agent-Editing World Model: Rethinking World Modeling for LLM Agents；MemBodied: Recurrent Associative Memory for Vision-Language-Action Models；IterSynth: Rethinking Deep Search Agents via Role-Decoupled Iterative Synthesis https://daily.jiayutool.cn/papers/2026-09-25/ https://daily.jiayutool.cn/papers/2026-09-25/ Fri, 25 Sep 2026 08:18:17 +0800 SpeakerMem-R1: Speaker-Centered Dual-Track Memory for Multi-Party Dialogue；Spatial-Interactor: Learning Spatial Reasoning through Interaction with the Observable Physical World；The Past Frames the Future: Memory for Autoregressive Video Generation；RewardVerse: Rubric-Guided Policy Optimization for Video Reward Modeling；From Pattern Recognizers to Personalized Companions: A Survey of Large Language Models in Mental Health；HappyWorld-Bench；Schrödinger's Code Repository: Have LLMs Learned SWE-bench or Memorized It?；Just-in-Time Memory: Learning to Curate Task-Adaptive Memory for LLM Agents https://daily.jiayutool.cn/papers/2026-09-24/ https://daily.jiayutool.cn/papers/2026-09-24/ Thu, 24 Sep 2026 08:11:11 +0800 GAE: Learning a Geometry-Native Latent Space for 3D-Consistent World Generation；Grounded Action Model: 3D Grounding as a Foundation for Robotics；The Tasteful Agent: Measuring and Improving Taste in Long-Horizon Tasks；Lean Pool: An AI-Maintained Archive of Formalized Mathematics；CARE: Experience-Guided Atomic Corrective Execution for Vision-Language-Action Policies；All-in-One Multilingual Scene Text Recognition with Script-aware Mixture-of-Experts；Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents；Ovis-Embedding: Pushing the Frontiers of Universal Omni-Modal Embeddings https://daily.jiayutool.cn/papers/2026-09-23/ https://daily.jiayutool.cn/papers/2026-09-23/ Wed, 23 Sep 2026 08:08:19 +0800 WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory；GameHorizon Suite: Multi-Horizon Data and Evaluation in Gameplay；RRSI: Regularized Recursive Self-Improvement of Agent Harnesses；OmniEdu: Open Foundation Models for Learning and Teaching；onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction；Realtime-Venus: A full-duplex interaction system with asynchronous delegation；Transferring the Intelligence of VLMs to Robotic Control；Harness-Zero: Harness Distillation via Agent-as-Harness https://daily.jiayutool.cn/papers/2026-09-22/ https://daily.jiayutool.cn/papers/2026-09-22/ Tue, 22 Sep 2026 08:09:37 +0800 EvoOntology: A Self-Evolving Ontology Layer for Data Agents；RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents；Grounded Skill Synthesis from Code at Scale for Agentic Intelligence；OmniVChat: Synthesizing, Benchmarking, and Training for Native Audio-Visual Dialogue；CodeMidas: Scaling Agentic Coding RL Environments from Code Itself；IntBMoE: Integrating Block-Level Conditioning into Expert Composition for Full-Participation Mixture-of-Experts；Designer-RSI: Evolving Procedural Memory from User Traffic for Agentic Graphic Design；BI-Agent and BI-Bench: Towards Automating End-to-End Business Intelligence https://daily.jiayutool.cn/papers/2026-09-21/ https://daily.jiayutool.cn/papers/2026-09-21/ Mon, 21 Sep 2026 08:17:22 +0800 Self-Evolving Search Index；What Does Privileged Information Add to On-Policy Self-Distillation?；FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations；RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning；Region-Level Policy Optimization for Fine-grained MLLM Perception；SpectralShift: Effective Context Window Extension of Gated DeltaNet via Spectral Reparameterization；Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling；PANORAMA: Panoptic Grounded Captioning via Mask Proposal Selection https://daily.jiayutool.cn/papers/2026-09-20/ https://daily.jiayutool.cn/papers/2026-09-20/ Sun, 20 Sep 2026 08:09:41 +0800 VABench: Measuring Embodied Spatial Intelligence through Visual Demonstrations, Active Perception, and Metric Control；Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents；Verifiable Social Reasoning for LLM Assistants；Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL；RiskChainBench: A Benchmark for Obfuscated Platform Message Restoration and Evidence-Grounded Web Investigation；PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?；EvolveTrade: Experience-Driven Policy Refinement for Self-Evolving LLM Trading Agents；EventEgoHands++: Event-based Egocentric 3D Hand Mesh Reconstruction with Real Dataset https://daily.jiayutool.cn/papers/2026-09-19/ https://daily.jiayutool.cn/papers/2026-09-19/ Sat, 19 Sep 2026 08:11:22 +0800 SoL-Pi: Recursively Scaling Auto-Research Loops for Efficient Agent Harness；An Empirical Study of Harness Design for Coding Agents；WeVisDoc: From Coverage to Capability for Robust End-to-End Document Parsing；Can MiniMax-H3 Reason About the Physical World? An Evaluation of Omni-Modal Generative Model；When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation；JEPA-Anything: Learning Predictive Models across Different Worlds；ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE Tasks；DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression https://daily.jiayutool.cn/papers/2026-09-18/ https://daily.jiayutool.cn/papers/2026-09-18/ Fri, 18 Sep 2026 08:10:25 +0800 LimiX-2: A Contextual Mechanism Network Towards General Structured-Data Intelligence；ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments；Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening；Agora: Git as Shared Memory for Collective AutoResearch；The Other Half of the Memory Wall: Serving 35B MoEs from SSD with Trained Routing Prediction；In-Context Robot Learning with VLM Agents；ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Models；Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents https://daily.jiayutool.cn/papers/2026-09-17/ https://daily.jiayutool.cn/papers/2026-09-17/ Thu, 17 Sep 2026 08:09:21 +0800 Continual Learning Mechanisms Compose for Long-Horizon Memorization；The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement；AI for Games in the Foundation Model Era；StepAudio 3 Realtime Technical Report；ScienceBuddy: Recursive-in-Recursive Self-Improvement for Interactive Scientific Agents；AlayaVista: Streaming World Modeling from Panoramic States to Perspective Video；HazardAuditor: From Executable Threats to Safer Computer-Use Agents；Kaininja: Extending Native 3D Generators to the Part Level https://daily.jiayutool.cn/papers/2026-09-16/ https://daily.jiayutool.cn/papers/2026-09-16/ Wed, 16 Sep 2026 08:08:17 +0800 Atria Dawn: The Dawn of Agentic Superintelligence；Vidu S2: Real-Time Interactive, Editable, and Spatial Video Generation；ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search；Dream-RSI: Recursive Self-Improvement through Evolving Worlds；RSIAgent: Autonomous Exploration for Recursive Self-improvement in New Environments；PhysBrain 1.5: From Vision-Language Models to Physical Foundation Models；Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work；LynnReal-Omni: Native multi-modal Video Generation for Agentic Visual Workflows https://daily.jiayutool.cn/papers/2026-09-11/ https://daily.jiayutool.cn/papers/2026-09-11/ Fri, 11 Sep 2026 08:24:57 +0800 Show-Harness: Just a VLM Agent Can Play Robots；Programmable World Model；AgentGrad: Intervention-guided Prompt Optimization for Multi Agent Systems；SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?；WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data；$Φ$-Bench: Can Large Language Models Engineer the Infrastructure That Powers Them?；What Did I Just Say? Self-Listening for Full-Duplex Speech Models；A Three-Layer Caching Architecture for Low-Latency LLM Web Search on Commodity CPU Hardware https://daily.jiayutool.cn/papers/2026-09-10/ https://daily.jiayutool.cn/papers/2026-09-10/ Thu, 10 Sep 2026 06:49:01 +0800 AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing；NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness；Omni Interaction Agent Technical Report；Miles v0.1: Production-Level Post-Training；OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining；Marigold V2: Revisiting Diffusion Transformers for Monocular Depth Estimation；SceneMosaic: Efficient and Diverse Simulation-Ready Scene Generation via Hybrid Agentic Layout Evolution；DriveZero: End-to-End Driving Beyond Human Demonstrations https://daily.jiayutool.cn/papers/2026-09-09/ https://daily.jiayutool.cn/papers/2026-09-09/ Wed, 09 Sep 2026 06:48:36 +0800 Unlocking Lossless Speedups in LLMs via Discrete Diffusion；FlowBalance: Verifier-Grounded Self-Improvement from On-Policy Reasoning Experience；ENEAS: Embedding-guided Neural Ensemble for Adaptive Segmentation；EmbodiedSkills: A Unified Framework for Orchestrating, Training, and Deploying VLA Agents；The Attention Triangle in Audio-Video Models；Enoki: Efficient Multi-Level Hallucination Detection；Causal Foundation Models；Group Adaptive Clipping Policy Optimization