The Year in AI Papers
The year/October 2025

Monthly collection

October 2025

The papers that drew the most attention during this month of AI research.

Papers
98
Citations
3.6K
Research labs
5

Explore the year

View all months →
Aug2025, 63 papersSep2025, 79 papersOct2025, current month, 98 papersNov2025, 73 papersDec2025, 66 papersJan2026, 86 papersFeb2026, 86 papersMar2026, 100 papersApr2026, 77 papersMay2026, 115 papersJun2026, 86 papersJul2026, 84 papersAug2026, 3 papers
1,016papers
curated and indexed

98 papers published in October 2025

01

arXiv (Cornell University)

Less is More: Recursive Reasoning with Tiny Networks

The paper introduces Tiny Recursive Model (TRM), a simplified recursive reasoning approach that outperforms the Hierarchical Reasoning Model (HRM) and many large language models on hard puzzle tasks. TRM uses a single tiny 2-layer network (7M parameters) that recursively improves its answer, eliminating the need for fixed-point theorems, hierarchical…

Alexia Jolicoeur-Martineau
Published
Oct 2025
Upvotes
518
Citations
4
Read summary Original
02

arXiv.org

Agent Learning via Early Experience

The paper introduces the 'early experience' paradigm for training language agents, which uses the agent's own actions and resulting future states as supervision without external reward signals. This addresses limitations of both supervised fine-tuning (SFT) on expert data, which is costly and generalizes poorly, and reinforcement learning (RL), which is…

Kai Zhang, Xiangchao Chen, Bo Liu, Tianci Xue, et al.
Published
Oct 2025
Upvotes
276
Citations
57
Read summary Original
03

arXiv.org

Scaling Latent Reasoning via Looped Language Models

The paper introduces Ouro, a family of pre-trained Looped Language Models (LoopLM) that integrate iterative computation in latent space during pre-training, using an entropy-regularized objective for adaptive depth allocation. Trained on 7.7T tokens, Ouro 1.4B and 2.6B models match or exceed the performance of 4B and 8B standard transformers, achieving…

Rui-Jie Zhu, Zixuan Wang, Kai Hua, Tianyu Zhang, et al.
Published
Oct 2025
Upvotes
234
Citations
104
Read summary Original
04

NVIDIA

QeRL: Beyond Efficiency -- Quantization-enhanced Reinforcement Learning for LLMs

QeRL is a framework that combines NVFP4 4-bit quantization with Low-Rank Adaptation (LoRA) to accelerate and reduce memory usage in reinforcement learning (RL) for large language models (LLMs). The authors find that quantization noise increases policy entropy, which enhances exploration during RL, contrary to its detrimental effect in supervised…

Wei Huang, Yi Ge, Shuai Yang, Yicheng Xiao, et al.
Published
Oct 2025
Upvotes
183
Citations
16
Read summary Original
05

arXiv.org

Concerto: Joint 2D-3D Self-Supervised Learning Emerges Spatial Representations

Concerto is a self-supervised learning framework that combines 3D intra-modal self-distillation with 2D-3D cross-modal joint embedding prediction to learn spatial representations from point clouds and images. It outperforms standalone 2D (DINOv2) and 3D (Sonata) self-supervised models by 14.2% and 4.8% respectively in linear probing for 3D scene…

Yujia Zhang, Xiaoyang Wu, Yixing Lao, Chengyao Wang, et al.
Published
Oct 2025
Upvotes
181
Citations
29
Read summary Original
06

arXiv.org

Diffusion Transformers with Representation Autoencoders

The paper introduces Representation Autoencoders (RAEs), which replace the traditional VAE in latent diffusion models with a frozen pretrained representation encoder (e.g., DINOv2, SigLIP2, MAE) paired with a trained decoder. RAEs provide both high-quality reconstructions and semantically rich latent spaces, outperforming SD-VAE in reconstruction FID…

Boyang Zheng, Nanye Ma, Shengbang Tong, Saining Xie
Published
Oct 2025
Upvotes
171
Citations
221
Read summary Original
07

Neural Information Processing Systems

A Theoretical Study on Bridging Internal Probability and Self-Consistency for LLM Reasoning

This paper introduces a theoretical framework for sampling-based test-time scaling in LLM reasoning, decomposing reasoning error into estimation error and model error. It analyzes self-consistency (SC) and perplexity (PPL), finding SC has linear estimation error convergence while PPL has exponential but suffers from high model error and degradation when…

Zhi Zhou, Yuhao Tan, Zenan Li, Yuan Yao, et al.
Published
Oct 2025
Upvotes
151
Citations
11
Read summary Original
08

arXiv.org

Spatial Forcing: Implicit Spatial Representation Alignment for Vision-language-action Model

The paper introduces Spatial Forcing (SF), a method to enhance the spatial awareness of Vision-Language-Action (VLA) models without explicit 3D inputs. VLA models, built on 2D-pretrained VLMs, lack 3D understanding, limiting their robotic manipulation performance. Existing solutions using depth sensors or point clouds face issues like sensor noise and data…

Fuhao Li, Wenxuan Song, Han Zhao, Jingbo Wang, et al.
Published
Oct 2025
Upvotes
149
Citations
93
Read summary Original
09

arXiv.org

D2E: Scaling Vision-Action Pretraining on Desktop Data for Transfer to Embodied AI

The paper introduces D2E, a framework that uses desktop interactions (screen, keyboard, mouse) as a pretraining substrate for embodied AI. It comprises three components: the OWA Toolkit for scalable data collection and compression (152x), the Generalist-IDM for pseudo-labeling YouTube videos via timestamp-based next-event prediction (NEP-tau), and VAPT for…

Suhwan Choi, Jaeyoon Jung, Haebin Seong, Minchan Kim, et al.
Published
Oct 2025
Upvotes
147
Citations
0
Read summary Original
10

arXiv.org

Robot Learning: A Tutorial

This tutorial by Capuano et al. (University of Oxford, Hugging Face) provides a comprehensive overview of modern robot learning, tracing the shift from classical, model-based control to data-driven, learning-based paradigms. It covers foundational concepts in classical robotics, including forward/inverse kinematics and their limitations, then introduces…

Francesco Capuano, Caroline Pascal, Adil Zouitine, Thomas Wolf, et al.
Published
Oct 2025
Upvotes
141
Citations
5
Read summary Original
11

Moonshot AI

Kimi Linear: An Expressive, Efficient Attention Architecture

Kimi Linear is a hybrid linear attention architecture that outperforms full attention under fair comparisons across short-context, long-context, and reinforcement learning (RL) scaling regimes. Its core, Kimi Delta Attention (KDA), extends Gated DeltaNet with a finer-grained, channel-wise gating mechanism, improving memory control and enabling efficient…

Kimi Team, Yu Zhang, Zongyu Lin, Xingcheng Yao, et al.
Published
Oct 2025
Upvotes
137
Citations
116
Read summary Original
12

arXiv.org

Agentic Context Engineering: Evolving Contexts for Self-Improving Language Models

The paper introduces ACE (Agentic Context Engineering), a framework for context adaptation in LLMs that treats contexts as evolving playbooks, accumulating and refining strategies through generation, reflection, and curation. ACE addresses two limitations of prior methods: brevity bias (over-compression of prompts) and context collapse (loss of detail…

Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, et al.
Published
Oct 2025
Upvotes
134
Citations
236
Read summary Original
13

arXiv.org

PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model

PaddleOCR-VL is a state-of-the-art, resource-efficient document parsing solution from Baidu's PaddlePaddle team. It uses a two-stage architecture: PP-DocLayoutV2 for layout analysis (detection, classification, reading order) and PaddleOCR-VL-0.9B, a compact vision-language model combining a NaViT-style dynamic resolution visual encoder with the…

Cheng Cui, Ting Sun, Suyin Liang, Tingquan Gao, et al.
Published
Oct 2025
Upvotes
129
Citations
69
Read summary Original
14

arXiv.org

Thinking with Camera: A Unified Multimodal Model for Camera-Centric Understanding and Generation

The paper introduces Puffin, a unified multimodal model that jointly performs camera-centric understanding and generation. It treats camera parameters as a first-class modality, introducing 'thinking with camera' to align spatially grounded visual cues with professional photographic terms. Puffin integrates language regression and diffusion-based…

Kang Liao, Size Wu, Zhonghua Wu, Linyi Jin, et al.
Published
Oct 2025
Upvotes
128
Citations
15
Read summary Original
15

arXiv.org

Apriel-1.5-15b-Thinker

Apriel-1.5-15B-Thinker is a 15-billion-parameter open-weights multimodal reasoning model developed by ServiceNow's SLAM Lab. It achieves frontier-level performance through a data-centric mid-training approach rather than scale, starting from Pixtral-12B. The methodology involves three stages: depth upscaling to expand reasoning capacity, staged continual…

Shruthan Radhakrishna, Aman Tiwari, Aanjaneya Shukla, Masoud Hashemi, et al.
Published
Oct 2025
Upvotes
125
Citations
2
Read summary Original
16

arXiv.org

Efficient Long-context Language Model Training by Core Attention Disaggregation

This paper introduces core attention disaggregation (CAD), a technique to improve long-context LLM training by separating the parameter-free softmax(QK^T)V computation (core attention, CA) from other model components and scheduling it on a dedicated pool of resources. CAD leverages two key properties: statelessness (CA has no trainable parameters) and…

Yonghao Zhuang, Junda Chen, Bo Pang, Yi Gu, et al.
Published
Oct 2025
Upvotes
124
Citations
3
Read summary Original
17

arXiv.org

ReCode: Unify Plan and Action for Universal Granularity Control

The paper introduces ReCode, a novel paradigm for LLM-based agents that unifies planning and action within a single code representation to achieve universal control of decision granularity. ReCode treats high-level plans as abstract placeholder functions, which are recursively decomposed into finer-grained sub-functions until reaching executable primitive…

Zhaoyang Yu, Jiayi Zhang, Huixue Su, Yufan Zhao, et al.
Published
Oct 2025
Upvotes
124
Citations
11
Read summary Original
18

arXiv.org

The End of Manual Decoding: Towards Truly End-to-End Language Models

The paper introduces AutoDeco, a novel architecture that makes LLM generation truly end-to-end by learning to predict token-level temperature and top-p values during decoding. This replaces manual hyperparameter tuning with a dynamic, self-regulating process. AutoDeco uses lightweight heads on top of frozen base models, trained with a differentiable soft…

Zhichao Wang, Dongyang Ma, Xinting Huang, Deng Cai, et al.
Published
Oct 2025
Upvotes
121
Citations
6
Read summary Original
19

arXiv.org

Paper2Video: Automatic Video Generation from Scientific Papers

The paper introduces Paper2Video, the first benchmark of 101 research papers paired with author-created presentation videos, slides, and speaker metadata, along with four evaluation metrics: Meta Similarity, PresentArena, PresentQuiz, and IP Memory. It also proposes PaperTalker, a multi-agent framework that generates presentation videos from papers,…

Zeyu Zhu, Kevin Qinghong Lin, Mike Zheng Shou
Published
Oct 2025
Upvotes
120
Citations
24
Read summary Original
20

Conference on Empirical Methods in Natural Language Processing

When Models Lie, We Learn: Multilingual Span-Level Hallucination Detection with PsiloQA

The paper introduces PsiloQA, a large-scale multilingual dataset for span-level hallucination detection in LLMs, covering 14 languages. It is built via an automated pipeline: generating QA pairs from Wikipedia using GPT-4o, eliciting answers from diverse LLMs without context, annotating hallucinated spans with GPT-4o, and filtering low-quality samples. The…

Elisei Rykov, Kseniia Petrushina, Maksim Savkin, Valerii Olisov, et al.
Published
Oct 2025
Upvotes
117
Citations
10
Read summary Original
21

arXiv.org

Emu3.5: Native Multimodal Models are World Learners

Emu3.5 is a large-scale multimodal world model from BAAI that natively predicts the next state across vision and language. It is pre-trained end-to-end with a unified next-token prediction objective on over 10 trillion tokens from interleaved vision-language data, primarily derived from internet videos. The model accepts and generates interleaved…

Yufeng Cui, Honghao Chen, Haoge Deng, Xu Huang, et al.
Published
Oct 2025
Upvotes
117
Citations
109
Read summary Original
22

arXiv.org

LightMem: Lightweight and Efficient Memory-Augmented Generation

LightMem is a lightweight and efficient memory-augmented generation system for Large Language Models (LLMs), presented at ICLR 2026. It addresses inefficiencies in existing memory systems by drawing inspiration from the Atkinson-Shiffrin human memory model, organizing memory into three stages: sensory memory (pre-compression and topic segmentation),…

Jizhan Fang, Xinle Deng, Haoming Xu, Ziyan Jiang, et al.
Published
Oct 2025
Upvotes
116
Citations
111
Read summary Original
23

arXiv.org

Every Attention Matters: An Efficient Hybrid Architecture for Long-Context Reasoning

This technical report introduces the Ring-linear model series, comprising Ring-mini-linear-2.0 (16B total, 1.6B active parameters) and Ring-flash-linear-2.0 (104B total, 7.4B active). Both use a hybrid architecture that combines linear and softmax attention to reduce I/O and computational costs in long-context scenarios, cutting inference cost to 1/10 of a…

Ling Team, Bin Han, Caizhi Tang, Chen Liang, et al.
Published
Oct 2025
Upvotes
116
Citations
9
Read summary Original
24

Research paper

There is No VAE: End-to-End Pixel-Space Generative Modeling via Self-Supervised Pre-training

The paper introduces EPG, a two-stage training framework for pixel-space diffusion and consistency models that closes the performance gap with latent-space methods. Stage one pre-trains an encoder using a contrastive loss and a representation consistency loss, aligning features of noisy images along deterministic sampling trajectories. Stage two fine-tunes…

Jiachen Lei, Keli Liu, Julius Berner, Haiming Yu, et al.
Published
Oct 2025
Upvotes
115
Citations
15
Read summary Original
25

arXiv.org

In-the-Flow Agentic System Optimization for Effective Planning and Tool Use

The paper introduces AGENTFLOW, a trainable agentic framework for planning and tool use that coordinates four specialized modules—planner, executor, verifier, and generator—through an evolving memory. Unlike monolithic tool-integrated reasoning models, AGENTFLOW optimizes its planner on-policy within the multi-turn loop. To address long-horizon credit…

Zhuofeng Li, Haoxiang Zhang, Seungju Han, Sheng Liu, et al.
Published
Oct 2025
Upvotes
113
Citations
51
Read summary Original
26

arXiv.org

DeepAnalyze: Agentic Large Language Models for Autonomous Data Science

DeepAnalyze-8B is the first agentic large language model designed for autonomous data science, capable of completing the entire pipeline from raw data sources to analyst-grade research reports. It addresses the limitations of previous workflow-based agents by enabling autonomous orchestration and adaptive optimization through a curriculum-based agentic…

Shaolei Zhang, Ju Fan, Meihao Fan, Guoliang Li, et al.
Published
Oct 2025
Upvotes
113
Citations
43
Read summary Original
27

arXiv.org

MM-HELIX: Boosting Multimodal Long-Chain Reflective Reasoning with Holistic Platform and Adaptive Hybrid Policy Optimization

The paper introduces MM-HELIX, a benchmark of 1,260 samples across 42 multimodal tasks (algorithms, graphs, puzzles, games) with five difficulty levels, designed to evaluate long-chain reflective reasoning in MLLMs. Evaluation of 23 models shows significant deficits; even GPT-5 achieves only 58.1% accuracy, and open-source models lag far behind. To improve…

Xiangyu Zhao, Junming Lin, Tianhao Liang, Yifan Zhou, et al.
Published
Oct 2025
Upvotes
110
Citations
3
Read summary Original
28

Neural Information Processing Systems

Scaling Language-Centric Omnimodal Representation Learning

This paper introduces LCO-EMB, a language-centric framework for multimodal representation learning that leverages the implicit cross-modal alignment in multimodal large language models (MLLMs). The authors show that text-only contrastive fine-tuning of the language decoder reduces anisotropy and improves kernel-level similarity across image, audio, and…

Chenghao Xiao, Hou Pong Chan, Hao Zhang, Weiwen Xu, et al.
Published
Oct 2025
Upvotes
109
Citations
15
Read summary Original
29

arXiv.org

Agentic Entropy-Balanced Policy Optimization

The paper introduces Agentic Entropy-Balanced Policy Optimization (AEPO), an agentic reinforcement learning algorithm for training multi-turn web agents. It addresses two entropy-driven challenges: 'High-Entropy Rollout Collapse' (over-branching on few trajectories due to consecutive high-entropy tool-call steps) and 'High-Entropy Token Gradient Clipping'…

Guanting Dong, Licheng Bao, Zhongyuan Wang, Kangzhi Zhao, et al.
Published
Oct 2025
Upvotes
109
Citations
26
Read summary Original
30

International Conference on Automated Software Engineering

LongCodeZip: Compress Long Context for Code Language Models

LongCodeZip is a training-free, model-agnostic, plug-and-play framework for compressing long code contexts for code LLMs. It uses a dual-stage strategy: coarse-grained compression selects function-level chunks ranked by conditional perplexity relative to the instruction, and fine-grained compression segments retained functions into blocks via…

Yuling Shi, Yichun Qian, Hongyu Zhang, Beijun Shen, et al.
Published
Oct 2025
Upvotes
108
Citations
39
Read summary Original
31

Proceedings of the 25th International Conference on Autonomous Agents and Multiagent Systems

Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization

The paper investigates how fine-tuning Vision-Language-Action (VLA) models on robotic tasks degrades the visual representations inherited from their base Vision-Language Models (VLMs). The authors show that naive action fine-tuning causes representation collapse and attention sink, leading to domain-specific forgetting of vision-language knowledge. They…

Nikita Kachaev, Mikhail Kolosov, Daniil Zelezetsky, Alexey K. Kovalev, et al.
Published
Oct 2025
Upvotes
107
Citations
28
Read summary Original
32

Research paper

Tongyi DeepResearch Technical Report

Tongyi DeepResearch is an open-source agentic large language model designed for long-horizon, deep information-seeking research tasks. It uses an end-to-end training framework combining agentic mid-training (two-stage continual pre-training) and agentic post-training (SFT and reinforcement learning), supported by a fully automated synthetic data pipeline…

Tongyi DeepResearch Team, Baixuan Li, Bo Zhang, Dingchu Zhang, et al.
Published
Oct 2025
Upvotes
105
Citations
18
Read summary Original
33

The Web Conference

DeepAgent: A General Reasoning Agent with Scalable Toolsets

DeepAgent is an end-to-end deep reasoning agent that unifies autonomous thinking, tool discovery, and action execution within a single reasoning process, addressing limitations of predefined workflow-based agents. It introduces an autonomous memory folding mechanism that compresses interaction history into structured episodic, working, and tool memories,…

Xiaoxi Li, Wenxiang Jiao, Jiarui Jin, Guanting Dong, et al.
Published
Oct 2025
Upvotes
103
Citations
63
Read summary Original
34

arXiv.org

InteractComp: Evaluating Search Agents With Ambiguous Queries

The paper introduces INTERACTCOMP, a benchmark for evaluating whether search agents can recognize and resolve ambiguous queries through interaction. It contains 210 expert-curated questions across 9 domains, built using a target-distractor methodology where questions use only shared attributes of a lesser-known target and a popular alternative, making them…

Mingyi Deng, Lijun Huang, Yani Fan, Fanqi Kong, et al.
Published
Oct 2025
Upvotes
100
Citations
11
Read summary Original
35

arXiv.org

Cache-to-Cache: Direct Semantic Communication Between Large Language Models

This paper introduces Cache-to-Cache (C2C), a new paradigm for direct semantic communication between large language models (LLMs) that bypasses text-based interaction. C2C projects and fuses the KV-cache of a source model (Sharer) into that of a target model (Receiver) using a neural cache fuser with a learnable gating mechanism. Oracle experiments showed…

Tianyu Fu, Zihan Min, Hanling Zhang, Jichao Yan, et al.
Published
Oct 2025
Upvotes
99
Citations
35
Read summary Original
36

arXiv.org

JanusCoder: Towards a Foundational Visual-Programmatic Interface for Code Intelligence

The paper introduces JANUSCODER, a suite of foundational models for multimodal code intelligence, establishing a unified visual-programmatic interface. It addresses the scarcity of high-quality multimodal code data by developing a data synthesis toolkit that leverages synergies between data modalities. This toolkit enables the creation of JANUSCODE-800K,…

Qiushi Sun, Jingyang Gong, Yang Liu, Qiaosheng Chen, et al.
Published
Oct 2025
Upvotes
99
Citations
7
Read summary Original
37

arXiv.org

Self-Forcing++: Towards Minute-Scale High-Quality Video Generation

Self-Forcing++ is a method for long-horizon video generation that extends autoregressive diffusion models beyond the training horizon of their teacher models. It addresses quality degradation from error accumulation by generating long self-rollouts (up to 100 seconds), re-injecting noise into these degraded sequences (backward noise initialization), and…

Justin Cui, Jie Wu, Ming Li, Tao Yang, et al.
Published
Oct 2025
Upvotes
98
Citations
156
Read summary Original
38

arXiv.org

DITING: A Multi-Agent Evaluation Framework for Benchmarking Web Novel Translation

The paper introduces DITING, the first comprehensive evaluation framework for web novel translation, assessing narrative and cultural fidelity across six dimensions: idiom translation, lexical ambiguity, terminology localization, tense consistency, zero-pronoun resolution, and cultural safety, supported by over 18K expert-annotated Chinese–English sentence…

Enze Zhang, Jiaying Wang, Mengxi Xiao, Jifei Liu, et al.
Published
Oct 2025
Upvotes
97
Citations
4
Read summary Original
39

DeepSeek

DeepSeek-OCR: Contexts Optical Compression

DeepSeek-OCR is a vision-language model that explores optical compression of long contexts by mapping text to images and decoding them back to text. It consists of DeepEncoder, a novel vision encoder with low activation and high compression, and a DeepSeek3B-MoE decoder. On the Fox benchmark, it achieves 97% OCR precision at compression ratios under 10x,…

Haoran Wei, Yaofeng Sun, Yukun Li
Published
Oct 2025
Upvotes
95
Citations
159
Read summary Original
40

NVIDIA

OmniVinci: Enhancing Architecture and Data for Omni-Modal Understanding LLM

OmniVinci is an open-source omni-modal LLM from NVIDIA that jointly understands vision, audio, and text. The paper introduces three architectural innovations: OmniAlignNet, which aligns vision and audio embeddings in a shared latent space via contrastive learning; Temporal Embedding Grouping (TEG), which organizes embeddings by timestamps to capture…

Hanrong Ye, Chao-Han Huck Yang, Arushi Goel, Wei Huang, et al.
Published
Oct 2025
Upvotes
93
Citations
50
Read summary Original
41

arXiv.org

GEM: A Gym for Agentic LLMs

GEM (General Experience Maker) is an open-source environment simulator for agentic LLMs, analogous to OpenAI-Gym for traditional RL. It provides a standardized environment-agent interface with reset() and step(), asynchronous vectorized execution, and modular wrappers. GEM includes diverse tasks (Math, Code, Game, QA, ReasoningGym, Terminal) and tools…

Zichen Liu, Anya Sims, Keyu Duan, Changyu Chen, et al.
Published
Oct 2025
Upvotes
92
Citations
11
Read summary Original
42

arXiv.org

ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning

ThinkMorph is a unified multimodal model fine-tuned on ~24K interleaved reasoning traces across four tasks (Jigsaw Assembly, Spatial Navigation, Visual Search, Chart Refocus) to enable complementary text-image chain-of-thought reasoning. It outperforms its base model Bagel-7B by an average of 34.74% on vision-centric benchmarks, with gains of 85.84% on…

Jiawei Gu, Yunzhuo Hao, Huichen Will Wang, Linjie Li, et al.
Published
Oct 2025
Upvotes
88
Citations
40
Read summary Original
43

arXiv.org

KORMo: Korean Open Reasoning Model for Everyone

This paper introduces KORMo-10B, a 10.8B-parameter fully open bilingual Korean-English language model trained from scratch on a corpus where 68.74% of the Korean portion is synthetic. The authors systematically investigate the feasibility of using synthetic data for non-English fully open models (FOMs), addressing stability, tokenizer design, and bias.…

Minjun Kim, Hyeonseok Lim, Hangyeol Yoo, Inho Won, et al.
Published
Oct 2025
Upvotes
87
Citations
4
Read summary Original
44

arXiv.org

Video-Thinker: Sparking "Thinking with Videos" via Reinforcement Learning

Video-Thinker is a framework that extends the 'Thinking with Images' paradigm to video reasoning by enabling Multimodal Large Language Models (MLLMs) to autonomously use intrinsic 'grounding' (temporal localization) and 'captioning' (visual comprehension) capabilities within chain-of-thought reasoning, eliminating the need for external tools. The authors…

Shijian Wang, Jiarui Jin, Xingjian Wang, Linxin Song, et al.
Published
Oct 2025
Upvotes
87
Citations
21
Read summary Original
45

arXiv.org

WithAnyone: Towards Controllable and ID Consistent Image Generation

The paper introduces WithAnyone, a diffusion-based model for controllable and identity-consistent image generation, addressing the 'copy-paste' artifact where models replicate reference faces instead of preserving identity across variations. The authors construct MultiID-2M, a large-scale dataset with 500k group photos and paired references for 3k…

Hengyuan Xu, Wei Cheng, Peng Xing, Yixiao Fang, et al.
Published
Oct 2025
Upvotes
86
Citations
14
Read summary Original
46

arXiv.org

BAPO: Stabilizing Off-Policy Reinforcement Learning for LLMs via Balanced Policy Optimization with Adaptive Clipping

This paper introduces BAPO (Balanced Policy Optimization with Adaptive Clipping), a method to stabilize off-policy reinforcement learning for large language models (LLMs). Off-policy RL, where stale data from past policies is used, suffers from unstable optimization, entropy collapse, and gradient explosions. The authors identify two key issues: (1) an…

Zhiheng Xi, Xin Guo, Yang Nan, Enyu Zhou, et al.
Published
Oct 2025
Upvotes
86
Citations
48
Read summary Original
47

Research paper

Every Activation Boosted: Scaling General Reasoner to 1 Trillion Open Language Foundation

Ling 2.0 is a series of reasoning-oriented language foundation models from Inclusion AI, scaling from 16B to 1T parameters under a unified high-sparsity Mixture-of-Experts (MoE) paradigm. The series includes three non-thinking instruct models: Ling-mini-2.0 (16B total, 1.4B activated), Ling-flash-2.0 (103B total, 6.1B activated), and Ling-1T (1T total, 51B…

Ling Team, Ang Li, Ben Liu, Binbin Hu, et al.
Published
Oct 2025
Upvotes
86
Citations
30
Read summary Original
48

arXiv.org

ExGRPO: Learning to Reason from Experience

The paper introduces ExGRPO, a framework for improving reinforcement learning from verifiable rewards (RLVR) for large reasoning models by reusing past rollout experiences. The authors first analyze what makes a reasoning experience valuable, identifying that medium-difficulty questions (with rollout correctness between 25% and 75%) and low-entropy…

Runzhe Zhan, Yafu Li, Zhi Wang, Xiaoye Qu, et al.
Published
Oct 2025
Upvotes
83
Citations
41
Read summary Original
49

arXiv.org

RAG-Anything: All-in-One RAG Framework

RAG-Anything is a unified framework for multimodal Retrieval-Augmented Generation (RAG) that addresses the limitation of existing text-only RAG systems in handling real-world documents containing text, images, tables, and equations. The framework introduces dual-graph construction, which builds a cross-modal knowledge graph for non-textual content and a…

Zirui Guo, Xubin Ren, Lingrui Xu, Jiahao Zhang, et al.
Published
Oct 2025
Upvotes
83
Citations
18
Read summary Original
50

arXiv.org

UniVideo: Unified Understanding, Generation, and Editing for Videos

UniVideo is a unified multimodal framework for video understanding, generation, and editing, presented at ICLR 2026. It uses a dual-stream architecture combining a frozen Multimodal Large Language Model (MLLM, Qwen2.5VL-7B) for understanding and a Multimodal DiT (MMDiT, HunyuanVideo-T2V-13B) for generation, connected via a trainable MLP connector. This…

Cong Wei, Quande Liu, Zixuan Ye, Qiulin Wang, et al.
Published
Oct 2025
Upvotes
81
Citations
68
Read summary Original
51

arXiv.org

FineVision: Open Data Is All You Need

FineVision is a large-scale, open corpus of 24 million samples (17 million images, 89 million turns, 9.5 billion answer tokens) for training vision-language models. It unifies over 200 public sources into 185 subsets via a semi-automated, human-in-the-loop pipeline that includes schema mapping, cleaning, de-duplication, and decontamination against 66…

Luis Wiedmann, Orr Zohar, Amir Mahla, Xiaohan Wang, et al.
Published
Oct 2025
Upvotes
81
Citations
38
Read summary Original
52

arXiv.org

INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formats

This paper systematically compares integer (INT) and floating-point (FP) low-bit quantization formats for LLMs across different granularities. The authors introduce a theoretical framework based on quantization signal-to-noise ratio (QSNR) and crest factor, revealing a performance crossover: FP excels at coarse granularities, but at fine-grained block-wise…

Mengzhao Chen, Meng Wu, Hui Jin, Zhihang Yuan, et al.
Published
Oct 2025
Upvotes
81
Citations
18
Read summary Original
53

arXiv.org

World-in-World: World Models in a Closed-Loop World

The paper introduces World-in-World, the first open benchmark for evaluating generative world models (WMs) in a closed-loop embodied setting, moving beyond visual quality to task success. It provides a unified online planning strategy and a standardized action API to integrate diverse WMs into four embodied tasks: Active Recognition, Image-Goal Navigation,…

Jiahan Zhang, Muqing Jiang, Nanru Dai, Taiming Lu, et al.
Published
Oct 2025
Upvotes
78
Citations
35
Read summary Original
54

arXiv.org

Ming-UniVision: Joint Image Understanding and Generation with a Unified Continuous Tokenizer

Ming-UniVision introduces MingTok, a continuous visual tokenizer with a three-stage architecture (low-level encoder, semantic decoder, pixel decoder) that unifies image understanding and generation in a single autoregressive framework. Unlike discrete tokenizers, MingTok avoids quantization errors, balancing compact latents for generation with…

Ziyuan Huang, DanDan Zheng, Cheng Zou, Rui Liu, et al.
Published
Oct 2025
Upvotes
77
Citations
26
Read summary Original
55

arXiv.org

AI for Service: Proactive Assistance with AI Glasses

The paper introduces AI for Service (AI4Service), a paradigm for proactive, real-time assistance via AI glasses, contrasting with reactive AI that only responds to explicit commands. It proposes Alpha-Service, a framework inspired by the von Neumann architecture, comprising five components: Input Unit (perception via dual MLLMs), Central Processing Unit…

Zichen Wen, Yiyu Wang, Chenfei Liao, Boxue Yang, et al.
Published
Oct 2025
Upvotes
77
Citations
16
Read summary Original
56

Annual Meeting of the Association for Computational Linguistics

Low-probability Tokens Sustain Exploration in Reinforcement Learning with Verifiable Reward

This paper addresses the exploration collapse in Reinforcement Learning with Verifiable Rewards (RLVR) for large language models, where performance plateaus as policy entropy collapses. The authors identify the systematic elimination of valuable low-probability exploratory tokens, termed 'reasoning sparks' (e.g., 'wait', 'however'), as a key cause, while…

Guanhua Huang, Tingqiang Xu, Mingze Wang, Qi Yi, et al.
Published
Oct 2025
Upvotes
76
Citations
23
Read summary Original
57

arXiv.org

DreamOmni2: Multimodal Instruction-based Editing and Generation

DreamOmni2 introduces two new tasks: multimodal instruction-based editing and generation, which accept both text and image instructions and support concrete objects as well as abstract attributes (e.g., texture, pose, style). The authors propose a three-stage data synthesis pipeline: (1) a feature mixing scheme in attention to generate paired images…

Bin Xia, Bohao Peng, Yuechen Zhang, Junjia Huang, et al.
Published
Oct 2025
Upvotes
74
Citations
34
Read summary Original
58

arXiv.org

FlashWorld: High-quality 3D Scene Generation within Seconds

FlashWorld is a generative model that creates 3D scenes from a single image or text prompt in seconds, being 10-100x faster than previous methods while achieving superior rendering quality. It shifts from the conventional multi-view-oriented (MV-oriented) paradigm to a 3D-oriented approach that directly produces 3D Gaussian representations during…

Xinyang Li, Tengfei Wang, Zixiao Gu, Shengchuan Zhang, et al.
Published
Oct 2025
Upvotes
74
Citations
26
Read summary Original
59

arXiv.org

Continuous Autoregressive Language Models

This paper introduces Continuous Autoregressive Language Models (CALM), a paradigm shift from discrete next-token prediction to continuous next-vector prediction. CALM uses a high-fidelity autoencoder to compress K tokens into a single continuous vector, reducing the number of autoregressive steps by a factor of K. The authors develop a likelihood-free…

Chenze Shao, Darren Li, Fandong Meng, Jie Zhou
Published
Oct 2025
Upvotes
74
Citations
10
Read summary Original
60

arXiv.org

Every Step Evolves: Scaling Reinforcement Learning for Trillion-Scale Thinking Model

Ring-1T is the first open-source, state-of-the-art thinking model with 1 trillion total parameters and about 50 billion activated per token, built on the Ling 2.0 architecture. Training at this scale posed challenges like train-inference misalignment, rollout inefficiencies, and RL system bottlenecks. The authors introduced three innovations: IcePop, which…

Ling Team, Anqi Shen, Baihui Li, Bin Hu, et al.
Published
Oct 2025
Upvotes
73
Citations
32
Read summary Original
61

Annual Meeting of the Association for Computational Linguistics

OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows

This paper introduces OS-Sentinel, a hybrid framework for detecting safety risks in mobile GUI agents powered by Vision-Language Models (VLMs). The authors first construct MobileRisk-Live, a dynamic Android emulator sandbox that records GUI observations, agent actions, and a System State Trace (T_sys) capturing underlying system metadata. From this, they…

Qiushi Sun, Mukai Li, Zhoumianze Liu, Zhihui Xie, et al.
Published
Oct 2025
Upvotes
73
Citations
15
Read summary Original
62

arXiv.org

AgentFold: Long-Horizon Web Agents with Proactive Context Management

AgentFold is a new web agent paradigm that addresses the trade-off between context saturation in ReAct-based agents and information loss from fixed summarization. It treats context as a dynamic workspace with Multi-Scale State Summaries and a Latest Interaction, using a 'folding' operation with two modes: Granular Condensation (preserving fine details) and…

Rui Ye, Zhongwang Zhang, Kuan Li, Huifeng Yin, et al.
Published
Oct 2025
Upvotes
73
Citations
65
Read summary Original
63

arXiv.org

From Pixels to Words -- Towards Native Vision-Language Primitives at Scale

This paper introduces NEO, a family of native Vision-Language Models (VLMs) that unify vision and language processing in a single monolithic architecture, contrasting with modular VLMs that combine separate visual encoders and language models. NEO is built from first principles using three key primitives: flexible position encoding, Multi-Head Native…

Haiwen Diao, Mingxuan Li, Silei Wu, Linjun Dai, et al.
Published
Oct 2025
Upvotes
70
Citations
12
Read summary Original
64

arXiv.org

Language Models are Injective and Hence Invertible

This paper proves that decoder-only Transformer language models are almost surely injective: distinct input prompts map to distinct last-token hidden representations, both at initialization and after any finite number of gradient descent steps. The authors establish this by showing the model is a real-analytic function of its parameters, so collisions can…

Giorgos Nikolaou, Tommaso Mencattini, Donato Crisostomi, Andrea Santilli, et al.
Published
Oct 2025
Upvotes
70
Citations
31
Read summary Original
65

Annual Meeting of the Association for Computational Linguistics

Human-Agent Collaborative Paper-to-Page Crafting

AutoPage is a multi-agent system that automatically converts academic papers into interactive project webpages. It uses a coarse-to-fine pipeline with three phases: narrative planning, multimodal content generation, and interactive page rendering. Dedicated 'Checker' agents verify each step against the source paper to prevent hallucination, and optional…

Qianli Ma, Siyu Wang, Yilin Chen, Yinhao Tang, et al.
Published
Oct 2025
Upvotes
70
Citations
6
Read summary Original
66

Volume 1

Glyph: Scaling Context Windows via Visual-Text Compression

Glyph is a framework that scales context windows by rendering long texts into images and processing them with vision-language models (VLMs), achieving 3-4x token compression while maintaining accuracy comparable to leading LLMs like Qwen3-8B. It consists of three stages: continual pre-training on rendered long-text data, an LLM-driven genetic search to…

Jiale Cheng, Yusen Liu, Xinyu Zhang, Yulin Fei, et al.
Published
Oct 2025
Upvotes
69
Citations
42
Read summary Original
67

arXiv.org

VLA-RFT: Vision-Language-Action Reinforcement Fine-tuning with Verified Rewards in World Simulators

VLA-RFT is a reinforcement fine-tuning framework for Vision-Language-Action (VLA) models that uses a data-driven world model as a simulator to provide verified rewards, avoiding costly real-world interactions and sim-to-real gaps. The world model, pretrained on offline data, predicts future visual observations conditioned on actions, enabling policy…

Hengtao Li, Pengxiang Ding, Runze Suo, Yihao Wang, et al.
Published
Oct 2025
Upvotes
68
Citations
50
Read summary Original
68

arXiv.org

UniGenBench++: A Unified Semantic Evaluation Benchmark for Text-to-Image Generation

UniGenBench++ is a unified semantic evaluation benchmark for text-to-image (T2I) generation, addressing limitations in existing benchmarks such as coarse evaluation dimensions, lack of prompt diversity, and missing multilingual support. It comprises 600 prompts hierarchically organized into 5 primary themes and 20 subthemes, covering 10 primary and 27…

Yibin Wang, Zhimin Li, Yuhang Zang, Jiazi Bu, et al.
Published
Oct 2025
Upvotes
68
Citations
20
Read summary Original
69

arXiv.org

TaTToo: Tool-Grounded Thinking PRM for Test-Time Scaling in Tabular Reasoning

TATTOO is a novel table-grounded Process Reward Model (PRM) designed to improve test-time scaling (TTS) for large reasoning models (LRMs) on tabular reasoning tasks. The authors identify that existing PRMs fail on table-specific operations like sub-table retrieval and schema interaction, leading to performance bottlenecks. TATTOO addresses this by…

Jiaru Zou, Soumya Roy, Vinay Kumar Verma, Ziyi Wang, et al.
Published
Oct 2025
Upvotes
67
Citations
10
Read summary Original
70

arXiv.org

A Survey of Data Agents: Emerging Paradigm or Overstated Hype?

This survey introduces the first systematic hierarchical taxonomy for data agents, inspired by the SAE J3016 standard for driving automation, to address the terminological ambiguity surrounding the term. The taxonomy comprises six levels (L0–L5) that delineate progressive shifts in autonomy, from manual operations (L0) to a vision of generative, fully…

Yizhang Zhu, Liangwei Wang, Chenyu Yang, Xiaotian Lin, et al.
Published
Oct 2025
Upvotes
67
Citations
36
Read summary Original
71

Research paper

$π_\texttt{RL}$: Online RL Fine-tuning for Flow-based Vision-Language-Action Models

The paper introduces π_RL, a framework for online reinforcement learning (RL) fine-tuning of flow-based Vision-Language-Action (VLA) models such as π0 and π0.5. It addresses the challenge of intractable action log-likelihoods in flow matching by proposing two methods: Flow-Noise, which models the denoising process as a discrete-time MDP with a learnable…

Kang Chen, Zhihao Liu, Tonghe Zhang, Zhen Guo, et al.
Published
Oct 2025
Upvotes
66
Citations
3
Read summary Original
72

arXiv.org

VideoCanvas: Unified Video Completion from Arbitrary Spatiotemporal Patches via In-Context Conditioning

VideoCanvas is a unified framework for arbitrary spatio-temporal video completion, where users specify patches at any spatial location and timestamp, and the model generates a coherent video. The paper identifies that causal video VAEs compress multiple frames into a single latent slot, creating temporal ambiguity, and that zero-padding in video mode…

Minghong Cai, Qiulin Wang, Zongli Ye, Wenze Liu, et al.
Published
Oct 2025
Upvotes
65
Citations
4
Read summary Original
73

arXiv.org

UniMoE-Audio: Unified Speech and Music Generation with Dynamic-Capacity MoE

UniMoE-Audio is a unified speech and music generation model that addresses task conflict and data imbalance through a Dynamic-Capacity Mixture-of-Experts (MoE) framework. The architecture introduces a Top-P routing strategy for dynamic expert allocation and a hybrid expert design with routed, shared, and null experts. To handle data imbalance, a…

Zhenyu Liu, Yunxin Li, Xuanyu Zhang, Qixun Teng, et al.
Published
Oct 2025
Upvotes
65
Citations
6
Read summary Original
74

arXiv.org

NANO3D: A Training-Free Approach for Efficient 3D Editing Without Masks

Nano3D is a training-free framework for 3D object editing that supports removal, addition, and replacement tasks without requiring masks. It integrates FlowEdit into the TRELLIS pipeline to perform localized edits guided by front-view renderings, and introduces region-aware merging strategies (Voxel/Slat-Merge) to preserve structural fidelity in unedited…

Junliang Ye, Shenghao Xie, Ruowen Zhao, Zhengyi Wang, et al.
Published
Oct 2025
Upvotes
65
Citations
27
Read summary Original
75

arXiv.org

PICABench: How Far Are We from Physically Realistic Image Editing?

PICABench is a benchmark for evaluating physical realism in image editing, addressing the gap where existing models and benchmarks focus on semantic fidelity but overlook physical effects like shadows, reflections, and state transitions. It categorizes physical consistency into three dimensions—Optics, Mechanics, and State Transition—spanning eight…

Yuandong Pu, Le Zhuo, Songhao Han, Jinbo Xing, et al.
Published
Oct 2025
Upvotes
65
Citations
10
Read summary Original
76

arXiv.org

The Principles of Diffusion Models

This book provides a comprehensive, mathematically grounded introduction to diffusion models, tracing their origins from three foundational perspectives: variational (VAEs to DDPMs), score-based (EBMs to NCSN and Score SDEs), and flow-based (Normalizing Flows to Flow Matching). It unifies these views through the Fokker-Planck equation and a common…

Chieh-Hsin Lai, Yang Song, Dongjun Kim, Yuki Mitsufuji, et al.
Published
Oct 2025
Upvotes
64
Citations
69
Read summary Original
77

arXiv.org

BitNet Distillation

BitDistill is a lightweight pipeline for fine-tuning full-precision LLMs (e.g., Qwen3) into 1.58-bit ternary precision for specific downstream tasks, achieving performance comparable to full-precision models while reducing memory by 10× and speeding up CPU inference by 2.65×. The method comprises three stages: (1) modeling refinement with SubLN modules to…

Xun Wu, Shaohan Huang, Wenhui Wang, Ting Song, et al.
Published
Oct 2025
Upvotes
63
Citations
2
Read summary Original
78

arXiv.org

LoongRL: Reinforcement Learning for Advanced Reasoning over Long Contexts

LoongRL is a data-driven reinforcement learning (RL) method for advanced long-context reasoning. It introduces KeyChain, a synthesis approach that transforms short multi-hop QA into high-difficulty long-context tasks by inserting UUID chains that hide the true question among distracting documents. Solving these tasks requires tracing the chain, identifying…

Siyuan Wang, Gaokai Zhang, Li Lyna Zhang, Ning Shang, et al.
Published
Oct 2025
Upvotes
63
Citations
21
Read summary Original
79

Neural Information Processing Systems

AdaSPEC: Selective Knowledge Distillation for Efficient Speculative Decoders

AdaSPEC is a novel knowledge distillation method for speculative decoding (SD) that improves draft model alignment with the target model by selectively filtering tokens during training. Unlike conventional KD that minimizes KL divergence across all tokens, AdaSPEC identifies and excludes 'hard' tokens that are difficult for the small draft model to learn,…

Yuezhou Hu, Jiaxin Guo, Xinyu Feng, Tuo Zhao
Published
Oct 2025
Upvotes
62
Citations
7
Read summary Original
80

arXiv.org

RoboOmni: Proactive Robot Manipulation in Omni-modal Context

RoboOmni introduces cross-modal contextual instructions, a new setting for robotic manipulation where robots infer user intent from speech, environmental sounds, and visual cues rather than explicit commands. The authors propose a Perceiver-Thinker-Talker-Executor framework built on end-to-end omni-modal LLMs, which fuses auditory and visual signals for…

Siyin Wang, Jinlan Fu, Feihong Liu, Xinzhe He, et al.
Published
Oct 2025
Upvotes
62
Citations
8
Read summary Original
81

IEEE International Conference on Computer Vision

StealthAttack: Robust 3D Gaussian Splatting Poisoning via Density-Guided Illusions

The paper introduces StealthAttack, the first data poisoning attack targeting 3D Gaussian Splatting (3DGS) to inject view-dependent illusory objects. The method uses Kernel Density Estimation (KDE) to identify low-density regions in the Gaussian point cloud, placing poison points along rays from the target (poisoned) view to make illusions visible only…

Bo-Hsu Ke, You-Zhe Xie, Yu-Lun Liu, Wei-Chen Chiu
Published
Oct 2025
Upvotes
61
Citations
6
Read summary Original
82

arXiv.org

Large Reasoning Models Learn Better Alignment from Flawed Thinking

Large reasoning models (LRMs) generate chain-of-thought (CoT) before answering but are easily biased by flawed reasoning prefills, leading to unsafe or overrefused outputs. The paper introduces RECAP (Robust Safety Alignment via Counter-Aligned Prefilling), a reinforcement learning (RL) post-training method that trains models to override flawed reasoning…

ShengYun Peng, Eric Smith, Ivan Evtimov, Song Jiang, et al.
Published
Oct 2025
Upvotes
60
Citations
10
Read summary Original
83

arXiv.org

Bee: A High-Quality Corpus and Full-Stack Suite to Unlock Advanced Fully Open MLLMs

The paper introduces Honey-Data-15M, a 15-million-sample SFT dataset for fully open multimodal large language models (MLLMs), built to address data noise and a lack of complex reasoning data. It also presents HoneyPipe, a transparent data curation pipeline built on the DataStudio framework, which uses rule-based and model-based filtering, dual-level…

Yi Zhang, Bolin Ni, Xin-Sheng Chen, Heng-Rui Zhang, et al.
Published
Oct 2025
Upvotes
60
Citations
25
Read summary Original
84

arXiv.org

Attention Illuminates LLM Reasoning: The Preplan-and-Anchor Rhythm Enables Fine-Grained Policy Optimization

This paper investigates how attention dynamics in LLMs can reveal reasoning patterns and improve reinforcement learning (RL) credit assignment. The authors analyze attention heads, classifying them as locally or globally focused. Locally focused heads show a sawtooth pattern near the diagonal, indicating phrasal chunks, while globally focused heads…

Yang Li, Zhichen Dong, Yuhan Sun, Weixun Wang, et al.
Published
Oct 2025
Upvotes
59
Citations
25
Read summary Original
85

arXiv.org

FARMER: Flow AutoRegressive Transformer over Pixels

FARMER is a novel end-to-end generative framework that unifies Normalizing Flows (NF) and Autoregressive (AR) models for tractable likelihood estimation and high-quality image synthesis directly from raw pixels. It employs an invertible autoregressive flow to transform images into latent sequences, whose distribution is modeled implicitly by an AR model.…

Guangting Zheng, Qinyu Zhao, Tao Yang, Fei Xiao, et al.
Published
Oct 2025
Upvotes
59
Citations
13
Read summary Original
86

arXiv.org

StockBench: Can LLM Agents Trade Stocks Profitably In Real-world Markets?

STOCKBENCH is a new benchmark for evaluating LLM agents in realistic, multi-month stock trading environments, using data from March to June 2025 to avoid contamination. Agents receive daily prices, fundamentals, and news, and make sequential buy, sell, or hold decisions. Performance is measured by cumulative return, maximum drawdown, and Sortino ratio.…

Yanxu Chen, Zijun Yao, Yantao Liu, Amy Xin, et al.
Published
Oct 2025
Upvotes
57
Citations
26
Read summary Original
87

AAAI Conference on Artificial Intelligence

ImagerySearch: Adaptive Test-Time Search for Video Generation Beyond Semantic Dependency Constraints

The paper introduces ImagerySearch, a test-time search strategy for text-to-video generation that adapts to prompts with long-distance semantic relationships, which are rare in training data and cause performance degradation in imaginative scenarios. ImagerySearch dynamically adjusts the inference search space (SaDSS) and reward function (AIR) based on the…

Meiqi Wu, Jiashu Zhu, Xiaokun Feng, Chubin Chen, et al.
Published
Oct 2025
Upvotes
57
Citations
12
Read summary Original
88

arXiv.org

Game-TARS: Pretrained Foundation Models for Scalable Generalist Multimodal Game Agents

Game-TARS is a generalist game agent from ByteDance Seed that uses a unified, scalable action space based on native keyboard and mouse inputs, enabling large-scale continual pre-training across OS, web, and simulation games. It was pre-trained on over 500B tokens, including game trajectories, cross-domain agent data, and general multimodal data. Key…

Zihao Wang, Xujing Li, Yining Ye, Junjie Fang, et al.
Published
Oct 2025
Upvotes
57
Citations
16
Read summary Original
89

arXiv.org

Open-o3-Video: Grounded Video Reasoning with Explicit Spatio-Temporal Evidence

Open-o3-Video is a non-agent framework that integrates explicit spatio-temporal evidence into video reasoning by highlighting key timestamps, objects, and bounding boxes. The authors construct two datasets, STGR-CoT-30k and STGR-RL-36k, combining existing temporal and spatial grounding resources with 5.9k newly annotated spatio-temporal samples. They adopt…

Jiahao Meng, Xiangtai Li, Haochen Wang, Yue Tan, et al.
Published
Oct 2025
Upvotes
56
Citations
43
Read summary Original
90

arXiv.org

Lumina-DiMOO: An Omni Diffusion Large Language Model for Multi-Modal Generation and Understanding

Lumina-DiMOO is an open-source, 8B-parameter unified multi-modal model based on a fully discrete diffusion architecture, built on the LLaDA-Base dLLM. It supports text-to-image generation at arbitrary resolutions, image-to-image tasks (editing, style transfer, subject-driven, controllable, multi-view), and image understanding. The model uses a unified…

Yi Xin, Qi Qin, Siqi Luo, Kaiwen Zhu, et al.
Published
Oct 2025
Upvotes
55
Citations
87
Read summary Original
91

arXiv.org

GigaBrain-0: A World Model-Powered Vision-Language-Action Model

GigaBrain-0 is a Vision-Language-Action (VLA) foundation model that uses world model-generated data to reduce reliance on costly real-world robot data and improve generalization. It employs a mixture-of-transformers architecture with a pretrained PaliGemma2 VLM and a Diffusion Transformer for action generation, enhanced by RGBD input modeling and embodied…

GigaBrain Team, Angen Ye, Boyuan Wang, Chaojun Ni, et al.
Published
Oct 2025
Upvotes
54
Citations
42
Read summary Original
92

arXiv.org

AutoPR: Let's Automate Your Academic Promotion!

The paper introduces AutoPR, a novel task for automatically generating promotional content from academic papers, along with PRBench, a benchmark of 512 paper-post pairs, and PRAgent, a multi-agent framework. PRAgent operates in three stages: content extraction, multi-agent synthesis, and platform-specific adaptation. Evaluations on PRBench show PRAgent…

Qiguang Chen, Zheng Yan, Mingda Yang, Libo Qin, et al.
Published
Oct 2025
Upvotes
53
Citations
4
Read summary Original
93

arXiv.org

StreamingVLM: Real-Time Understanding for Infinite Video Streams

StreamingVLM is a framework for real-time understanding of infinite video streams, addressing the limitations of full attention (quadratic cost, poor long-video performance) and sliding window methods (coherence loss or high latency). It maintains a compact KV cache with attention sinks, a short vision window, and a long text window, using contiguous RoPE…

Ruyi Xu, Guangxuan Xiao, Yukang Chen, Liuning He, et al.
Published
Oct 2025
Upvotes
53
Citations
73
Read summary Original
94

arXiv.org

Latent Refinement Decoding: Enhancing Diffusion-Based Language Models by Refining Belief States

The paper introduces Latent Refinement Decoding (LRD), a two-stage decoding framework for diffusion-based language models (dLLMs) that addresses information loss and premature commitment in existing hard-assignment methods. LRD operates in two phases: Phase 1 (Latent Refinement) performs distribution-preserving updates in embedding space by mixing [MASK]…

Qinglin Zhu, Yizhen Yao, Runcong Zhao, Yanzheng Xiang, et al.
Published
Oct 2025
Upvotes
53
Citations
9
Read summary Original
95

arXiv.org

Detect Anything via Next Point Prediction

Rex-Omni is a 3B-parameter multimodal large language model (MLLM) that unifies object detection and other visual perception tasks under a next-point prediction framework. It uses quantized relative coordinates represented by 1,000 special tokens, reducing token usage per bounding box to four. Training combines supervised fine-tuning on 22 million samples…

Qing Jiang, Junan Huo, Xingyu Chen, Yuda Xiong, et al.
Published
Oct 2025
Upvotes
53
Citations
60
Read summary Original
96

arXiv.org

Chem-R: Learning to Reason as a Chemist

Chem-R is a chemical reasoning model designed to emulate the deliberative processes of chemists, addressing three key challenges in current LLMs: lack of chemical fundamentals, unreliable reasoning trajectories, and imbalanced performance across tasks. The model is trained via a three-phase framework: (1) Chemical Foundation Training, which fine-tunes on…

Weida Wang, Benteng Chen, Di Zhang, Wanhao Liu, et al.
Published
Oct 2025
Upvotes
53
Citations
11
Read summary Original
97

OpenAI

GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks

This paper introduces GDPval, a benchmark for evaluating AI models on real-world, economically valuable tasks. It covers 44 occupations across the top 9 U.S. GDP sectors, with tasks created by industry experts averaging 14 years of experience. The benchmark includes 1,320 tasks in the full set and a 220-task gold subset, graded via human expert pairwise…

Tejal Patwardhan, Rachel Dias, Elizabeth Proehl, Grace Kim, et al.
Published
Oct 2025
Upvotes
0
Citations
108
Read summary Original
98

Anthropic

Poisoning Attacks on LLMs Require a Near-constant Number of Poison Samples

This paper investigates whether poisoning attacks on large language models (LLMs) require a constant number of poisoned samples regardless of dataset size, rather than a fixed percentage. The authors conducted the largest pretraining poisoning experiments to date, training models from 600M to 13B parameters on Chinchilla-optimal datasets (6B to 260B…

Alexandra Souly, Javier Rando, Ed Chapman, Xander Davies, et al.
Published
Oct 2025
Upvotes
0
Citations
76
Read summary Original