ARCHIVE · 2026-10 · 198 PAPERS · PAGE 1/4

Papers of 2026-10

Papers first published on arXiv in 2026-10, newest first. To filter by topic, discipline or organisation, use the archive search.

Agents 83Visual generation 50Efficient training & inference 45Benchmarks & data 40Reinforcement learning 31Multimodal 28Robotics & embodied AI 26LLMs 20Reasoning 18Code & SWE 15Vision & 3D 14AI for Science 9Safety & interpretability 7Alignment 2

Top organisations: alibaba (5) · Tencent Hunyuan (5) · NVIDIA (4) · Nanyang Technological University (4) · Qwen (4)

▲182 ❝0 2026-10-20

Matrix-Game 4.0: Harnessing Video Models for Real-Time Playable Game Worlds

Matrix-Game 4.0 separates game logic (engine plus coding agent) from visuals (autoregressive video renderer), reaching playable 60 FPS on one RTX 5090.

Affiliation unverified · Jihai Zhang, Runjia Qian, Kaiqin Kong +15

Visual generation Agents arXiv HF

▲148 ❝0 2026-10-20

PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking

PAR aligns each loop iteration of a looped LM with task progress and anchors state norms, extrapolating state tracking to 16x training length.

ant-international · Pan Wang, Yihao Hu, Guangyuan Dong +8

LLMs Reasoning arXiv HF

▲94 ★12 ❝0 2026-10-20

S-Agent 2.0: Spatial Reasoning with a General Coding Agent

S-Agent 2.0 equips a general coding agent with command-line 3D perception tools and scene memory, lifting spatial-reasoning benchmarks.

Ropedia · Rui Gong, Ruoyu Wang, Yalun Dai +9

Agents Reasoning arXiv Code HF

▲64 ★574 ❝0 2026-10-20

Sage-A1.5: From Long-Horizon Execution to Scientific Exploration with Capability Composition

Sage-A1.5 is a 35B MoE agentic model adding subagent orchestration, self-verification and open-ended scientific exploration.

Intern Science · Wenjie Lou, Tianshuo Peng, Shiyang Feng +96

Agents AI for Science arXiv Code HF

▲47 ★215 ❝1 2026-10-20

SolarRenderer: Few-Step Generative Reference Rendering

SolarRenderer adds two LoRAs to a frozen video model to render coarse structural video into realistic footage in four steps, including real-time streaming.

Show Lab · Guian Fang, Shengju Qian, Dongyue Lu +6

Visual generation Efficient training & inference arXiv Code HF

▲42 ★17 ❝0 2026-10-20

Learning the 4D World through Point Trajectory Forecasting

VPM is a world model that jointly forecasts multi-view future video and 3D point trajectories, extended to robot control.

Affiliation unverified · Byungwoo Jeon, Sunghwan Hong, Dohyeon Kim +10

Visual generation Robotics & embodied AI arXiv Code HF

▲24 ★11 ❝0 2026-10-20

MMEnterprise-Bench: Evaluating Enterprise Agents over Evolving Multimodal Records

MMEnterprise-Bench has 1,613 questions over evolving multimodal workplace records, testing evidence reasoning, conflict resolution and unanswerability.

Affiliation unverified · Seohyun Lee, Jongha Kim, Dohwan Ko +4

Benchmarks & data Agents arXiv Code HF

▲22 ❝0 2026-10-20

Joints, Camera, Action! Co-Generating 3D Motion and Video for Arbitrary Rigs

Co-generates video, replayable 3D motion for any rig and a registering camera, turning video-model motion into usable 3D animation.

NVIDIA · Noam Rotstein, Yoad Tewel, Chen Tessler +5

Visual generation Vision & 3D arXiv HF

▲237 ❝0 2026-10-19

Scaling Parameter-Efficient Fine-Tuning for Brain Encoding

Uses parameter-efficient fine-tuning to adapt multimodal backbones of up to 104B for brain encoding, better predicting neural responses.

Mind Lab · Zhuofan Josh Ying, Guanglong Sun, Nolan Ho +6

AI for Science Efficient training & inference arXiv HF

▲225 ★12 ❝0 2026-10-19

Focused Omni-Modal GUI Agents

FocusGUI is a lightweight adapter that picks the step-relevant omni-modal evidence for a frozen GUI agent, curbing compounding distraction.

Nanyang Technological University · Fuyao Zhang, Wenwen He, Jiaming Zhang +6

Agents Multimodal arXiv Code HF

▲179 ★36 ❝0 2026-10-19

Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience

Ropedia W0 is an egocentric loco-manipulation world foundation model spanning humans and robots, trained on 10k hours of human and 3k hours of robot data.

Ropedia · Tianqi Liu, Hanfei Zhao, Qifeng Liang +12

Robotics & embodied AI Visual generation arXiv Code HF

▲170 ★18 2026-10-19

EgoXperience: Learning World-Grounded Egocentric Representations from Human Experience

EgoXperience jointly reconstructs observer motion, scene geometry and hand-object interaction from egocentric video, with an auto-annotation engine and benchmark.

Ropedia · Xinhao Ji, Hui En Pang, Hao Li +12

Vision & 3D Robotics & embodied AI Benchmarks & data arXiv Code HF

▲156 ★13 ❝0 2026-10-19

Holi-World: Building Holistic World Intelligence from Human-Ego Experience

Holi-World is an ecosystem for learning world intelligence from human egocentric experience, from capture to evaluation.

Ropedia · Hao Li, Xiyuan Zheng, Yuanyuan Gao +7

Vision & 3D Benchmarks & data arXiv Code HF

▲119 ★272 ❝0 2026-10-19

MemT-1.0: Towards Memory-Native Evolution

MemT-1.0 makes memory intrinsic to the model using Gated DeltaNet-2 and Mixture-of-Storage to scale native memory.

MemTensor · Zhiyu Li, Zeyu Zhang, Linyun Liu +71

LLMs Efficient training & inference arXiv Code HF

▲112 2026-10-19

Learning from Fleet Experience: Collective Experience Learning Across 96 Robots

Collective experience learning lets 96 deployed robots share execution data, retrieving successful segments for failures to fine-tune a common policy without new demos.

Affiliation unverified · Chao Yu, Shu'ang Yu, Hao Lin +41

Robotics & embodied AI arXiv HF

▲102 ★152 2026-10-19

OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory

OpenVLHarness gives MLLMs a tool capability layer and persistent multimodal memory for visual tasks needing external evidence.

Apple · Yu Zhou, Cheng-Fu Yang, Rui Sun +8

Multimodal Agents arXiv Code HF

▲84 ★40 2026-10-19

ScaleNav: Scalable Internet Video Pretraining for Language-Conditioned Local Navigation

ScaleNav learns goal-directed visual dynamics from internet egocentric videos for language-conditioned robot navigation.

Affiliation unverified · Yihui Yang, Yuanhong Yu, Jiawei Wang +7

Robotics & embodied AI arXiv Code HF

▲80 ★6 2026-10-19

SpecMRL: Redesign Matryoshka Representation Learning within Parametric Spectrum

SpecMRL reorganises features on a parameter-derived spectral basis so Matryoshka prefixes carry the most important information.

Affiliation unverified · Zhongxin Chen, Songhao Wu, Heng Cui +3

Efficient training & inference arXiv Code HF

▲37 ★7 2026-10-19

AnyGoal: On World Model Generalization Beyond Expert Rollouts

AnyGoal benchmarks whether world models generalize to recombinations of familiar state and action factors; current models consistently fail.

galilai-group · Heejeong Nam, Divyat Mahajan, Quentin Le Lidec +2

Benchmarks & data Agents arXiv Code HF

▲28 2026-10-19

Kronecker Delta Rule for Long Context Linear Attention

KronLSTM introduces a Kronecker delta rule that revises matrix memory along one direction, giving stable long-context linear attention that often beats DeltaNet.

NX-AI · Anamaria-Roberta Hartl, Levente Zólyomi, Michael List +3

Efficient training & inference LLMs arXiv HF

▲28 ★3 ❝0 2026-10-19

AdaExploreBench: Benchmarking LLM Agents for Proactive Exploration and Test-time Adaptation in Evolving Environments

AdaExploreBench uses hidden, silently drifting rules to test whether LLM agents proactively explore and adapt at test time.

HKUST · Rui Wang, Zishen Lai, Jiayu Liu +15

Benchmarks & data Agents arXiv Code HF

▲27 ★4 ❝0 2026-10-19

HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs

HarnessPro evolves agent harnesses without ground-truth tests, returning lite, balanced and heavy variants that trade performance against cost.

Affiliation unverified · Zhuoshi Pan, ZhiYuan Feng, Yukuo Cen +6

Agents Code & SWE arXiv Code HF

▲27 ★8 2026-10-19

RoAR-VAE: Improving Latent Diffusability through Robust Prediction

RoAR-VAE trains tokenizers so semantics and pixels remain recoverable from corrupted latents, making latents easier for diffusion models to learn.

Tsinghua-LeapLab · Yang Yue, Fangyun Wei, Yitong Wang +4

Visual generation arXiv Code HF

▲27 ❝0 2026-10-19

SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation

SWIFT extrapolates training targets away from a frozen weak denoiser to boost a stronger diffusion model with no extra inference cost.

alibaba · Yanjin He, Cuifeng Shen, Wanzhou Lei +5

Visual generation arXiv HF

▲26 ★31 ❝0 2026-10-19

InfiniDream: Open-Ended Photorealistic 3D Worlds Through Procedural Generation

InfiniDream has a VLM drive procedural algorithms to generate open-ended, editable photorealistic 3D assets and large worlds.

Alaya Lab · Yi-Chuan Huang, Hao-Jen Chien, Zheng-Hui Huang +3

Vision & 3D Code & SWE arXiv Code HF

▲26 ★47 ❝0 2026-10-19

Manga2Animate: An Agentic Framework for Faithful Manga Animation with Script and Storyboard Planning

Manga2Animate uses paired creator and reviewer agents to plan scripts and storyboards before generating animation faithful to the source manga.

National Yang Ming Chiao Tung University · Sheng-Wei Zhang, Hsi-An Chen, Yan-Ting Liou +3

Agents Visual generation arXiv Code HF

▲22 2026-10-19

SpecKCast: Spectral Kalman Forecasting for Diffusion Acceleration

SpecKCast forecasts DiT outputs in the frequency domain with Kalman filters, running the model only when uncertainty is high, for training-free speedup.

Samsung Research · Javad Rajabi, Hakki Can Karaimer, David B. Lindell +2

Visual generation Efficient training & inference arXiv HF

▲68 ★5 ❝0 2026-10-18

CoEvo-RSI: When Harnesses Evolve, Which Models Should Follow?

CoEvo-RSI tackles the brain–body mismatch when models and agent harnesses co-evolve, using a meta-agent to coordinate harness evolution and checkpoint choice.

Affiliation unverified · Jiaqi Su, Guibin Zhang, Jie-Jing Shao +5

Agents LLMs arXiv Code HF

▲60 ★9 ❝0 2026-10-18

VLA-Rolling: Fewer Denoising Steps, Better Control via Adaptive Action Reuse

VLA-Rolling reuses the unexecuted action tail as the starting point for denoising, cutting steps for frozen flow-matching VLAs while often raising success.

ZJU-OmniAI · Sicheng He, Yi Pan, Qiuting Yu +10

Robotics & embodied AI Efficient training & inference arXiv Code HF

▲42 ★12 ❝0 2026-10-18

CoworkEnv: Synthesizing Training Environments for Long-Horizon Cowork Agents

CoworkEnv synthesizes executable training environments for AI coworker agents, supporting SFT and RL of compact models.

Affiliation unverified · Zhongwei Xie, Yiming Du, Jiaxin Bai +10

Agents Reinforcement learning arXiv Code HF

▲37 ★22 2026-10-18

MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion

MT Forcing trains autoregressive-diffusion editors with a single-turn distillation teacher and teacher-forcing guidance to curb multi-turn editing drift.

SenseNova · Houyuan Chen, Fanzhou Wang, Xiangyu Fan +10

Visual generation arXiv Code HF

▲313 ★84 ❝0 2026-10-16

HexaShell: A Thin Harness for Exploring the Limits of VLM Agents as Robot Coding Agents

HexaShell keeps the harness thin so a coding agent drives a two-arm robot via a command line and writes its own tools after failures.

HexaFuture · Hongcheng Gao, Alex Zhou, Jay Zhu +13

Robotics & embodied AI Agents Code & SWE arXiv Code HF

▲291 ★5 ❝0 2026-10-16

OR-License: Evaluating LLM Agents as Real-World Operations Research Practitioners

OR-License benchmarks whether LLM agents can turn original client materials from 31 paid operations-research engagements into expert-grade solutions.

Data Science & Artificial Intelligence Lab at HKUST · Peisong Wang, Feixue Liu, Yize Meng +16

Benchmarks & data Agents arXiv Code HF

▲89 ★1.2k ❝0 2026-10-16

MOSS-Image Technical Report

MOSS-Image is an open 5B text-to-image system strong at bilingual long-text rendering and layouts, with intent-aware prompt enhancement.

OpenMOSS · MOSI. AI

Visual generation arXiv Code HF

▲62 ★27 ❝0 2026-10-16

Scene-Designer: Learning Design Intelligence for Agentic 3D Scene Generation

Scene-Designer learns design intelligence from design and critique experience to generate coherent 3D indoor scenes from text.

ZJU-OmniAI · Xuanyu Liu, Hongyan Feng, Miao Pan +5

Agents Vision & 3D arXiv Code HF

▲52 ❝0 2026-10-16

Tree-Aware Posterior Distillation for Long-Horizon Agentic Reinforcement Learning

TreeOPSD builds a tree over agent turns, backs up terminal rewards into turn-level credit and self-distills without a critic for long-horizon agent RL.

Tencent Hunyuan · Heming Zou, Qi Wang, Yixiu Mao +7

Reinforcement learning Agents arXiv HF

▲45 ★45 ❝0 2026-10-16

Xiaomi-Lens-0 Technical Report

Xiaomi-Lens-0 is a 4B MLLM on Qwen3.5 that predicts segmentation masks from its own visual representations.

SeerRay-Lab · Pei Fu, Yike Liu, Yue Sun +13

Vision & 3D Multimodal arXiv Code HF

▲45 ★28 ❝0 2026-10-16

Self-Evolving Coding Agents via Harness-Policy Optimization

Microsoft's Self-HPO alternately evolves a coding agent's harness and policy model for recursive self-improvement.

Microsoft Research · Zhepei Wei, Wenlin Yao, Baolin Peng +5

Code & SWE Agents Reinforcement learning arXiv Code HF

▲35 ★29 ❝0 2026-10-16

RAP: A Dataset for Paper-to-Poster Generation via Declarative Markup from Human-Crafted Exemplars

RAP pairs 25K+ papers with posters derendered into HTML/CSS slots, enabling render-and-compare evaluation of paper-to-poster generation.

meituan · Haonan Han, Yukang Lin, Minliang Lin +7

Benchmarks & data Code & SWE arXiv Code HF

▲33 ❝0 2026-10-16

MTEdit: Packed Trajectories Beat Rollouts for Multi-Turn Image Editing

MTEdit builds 500K multi-turn image-editing trajectories by generating several edit states in one packed canvas, plus a benchmark.

SenseNova · Houyuan Chen, Fanzhou Wang, Xiangyu Fan +8

Visual generation Benchmarks & data arXiv HF

▲33 ★12 ❝0 2026-10-16

Execution Before Memory: Diagnosing Memory Failures in Video World Models

MIST separates instruction execution from memory in video world model tests, finding the bottleneck is usually execution rather than forgetting.

The University of Hong Kong · Weiyang Jin, Jiwen Yu, Sihan Xu +2

Benchmarks & data Visual generation arXiv Code HF

▲32 ★43 ❝0 2026-10-16

DLite: History-free Speculative Decoding for Agentic AI

DLite is history-free speculative decoding that drafts locally from the target model's hidden states, suited to ever-growing agent contexts.

Shanghai Jiao Tong University · Jiaming Xu, Hanzhen Wang, Tianlang Zhao +7

Efficient training & inference Agents arXiv Code HF

▲727 ★5 ❝0 2026-10-15

PremiseCheck: Efficient Fact-Checking of Human and AI Peer Reviews

ReviewFact labels factual errors in peer reviews; PremiseCheck combines an LLM with cheap decision models to verify them efficiently.

Affiliation unverified · Dingjie Song, Zhiling Yan, Ziqian Zhang +2

Benchmarks & data Efficient training & inference arXiv Code HF

▲560 ★17 ❝0 2026-10-15

MJev-Spatial: System One Spatial Intelligence via Native Attention Matching

MJev-Spatial casts spatial questions as classification over candidates, answering relation, distance and point queries faster than text generation.

Mind Lab · Luozheng Qin, Yuhan Zhu, Changhai Zhou +9

Multimodal Efficient training & inference arXiv Code HF

▲459 ★218 ❝0 2026-10-15

Has GPT-6-Astra Achieved Embodied Intelligence?

Evaluates GPT-6-Astra on 12 robotics benchmarks and discusses what a general model's embodied competence means for the field.

SenseNova · SenseNova, Joint Research Team, Zhongang Cai +59

Robotics & embodied AI Benchmarks & data arXiv Code HF

▲330 ★396 ❝0 2026-10-15

OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use

OpenGCUA aims to open the full stack for generalist computer-use agents, with a first release spanning GUI and code execution.

Tongyi-MAI · Hanzhang Zhou, Long Li, Quyu Kong +12

Agents Code & SWE arXiv Code HF

▲272 ★107 ❝0 2026-10-15

FDE-Gym: The Last Mile of AI Delivery

FDE-Gym offers 95 real delivery tasks testing whether agents can clarify vague requirements and ship working solutions like forward-deployed engineers.

DeepLethe · Dongxu Yang, Hongda Yan, Renxin Wang +23

Agents Benchmarks & data arXiv Code HF

▲218 ★10 ❝0 2026-10-15

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

Waypoint-OPD uses an intermediate teacher checkpoint to transfer only consistent RL-induced policy shifts to the student.

Affiliation unverified · Bingxi Hou, Guochao Jiang, Guofeng Quan +4

Reinforcement learning Efficient training & inference arXiv Code HF

▲175 ❝0 2026-10-15

Learning to Think and Explore in 3D World

Trains VLMs with RL to reason and actively move their viewpoint in 3D, seeing one image at a time like a human.

University of California at San Diego · Yijiang Li, Bingyang Wang, Nuno Vasconcelos

Multimodal Reinforcement learning Reasoning arXiv HF

▲117 ★185 ❝0 2026-10-15

3DCodeVerse: Exploring 3D Worlds as Code

3DCodeVerse offers 2M executable 3D programs and a coding harness so agents can create editable, verifiable 3D content.

University of Southern California · Yipeng Gao, Ziyao Zeng, Jinfa Huang +16

Code & SWE Visual generation arXiv Code HF