Research

• 223WeVisOne-VL: Unified Visual Perception from Objects to PixelsWeVisOne-VL unifies object-level and pixel-level visual perception via structured token prediction, supporting detection, localization, counting, OCR, segmentation, and matting with a lightweight pixel decoder and multi-stage training.Oct 21, 2026
• 165Making MLLM Latents Matter: Factor-Aware Predictive Learning for Cinematic Video GenerationFactor-Aware Predictive Learning enables video diffusion transformers to leverage multimodal language model representations for sparse temporal inference and factor-aware cinematic control through staged grounding and optimization.Oct 22, 2026
• 152Deep Search in Real-World Web EnvironmentsOpenDS-World introduces a live-web benchmark to evaluate autonomous multi-step deep search agents, revealing significant gaps between frontier models and human performance.Oct 18, 2026
• 74Editing Images into Reusable Layers via Compositional Edit-Prior DistillationLayerEditor unifies instruction-guided image editing with reusable RGBA layer generation via joint token modeling and edit-prior distillation, enabling precise local manipulation and layer reuse.Oct 16, 2026
• 55IQuest-Q1: Advancing Foundation Capabilities for Agentic CLI SystemsIQuest-Q1 is a sparse Mixture-of-Experts foundation model trained via multi-stage pre-training, mid-training, and post-training to achieve strong agentic coding and general capabilities.Oct 21, 2026
• 47Toward Recursive Self-Improvement in Physical AIOct 20, 2026
• 43Policy-Mask: Budget-Aware Adaptation of Diffusion Features to Discriminative TasksPolicy-Mask learns a budget-conditioned selection policy over diffusion timesteps, network blocks, and channels to extract compact, task-optimal features for discriminative vision tasks.Oct 19, 2026
• 39Deep Fusion Memory: A Chunk-Level Parametric Memory for Large Language ModelsDeep Fusion Memory integrates predicted parametric memory into intermediate hidden states via gated cross-attention, reducing retrieval overhead and improving knowledge-intensive tasks without online retrieval.Oct 17, 2026
• 32What Generation Knows: Unlocking Fine-Grained Understanding in Unified Multimodal ModelsGAFU leverages native generation-side representations to improve fine-grained visual understanding in unified multimodal models by compressing them into compact tokens for task-conditioned retrieval.Oct 13, 2026
• 29LaLa-World: Scaling Latent World Models with Natural Language ActionsLaLa-World is a language-conditioned latent world model that jointly learns from diverse simulation, robot, and video data to enable efficient model-predictive planning and policy initialization.Oct 20, 2026
• 25Meta Expert: Cross-Layer Expert Sharing for Storage- and Retrieval-Efficient MoEShared meta-experts across layers reduce redundancy and improve efficiency compared to layer-specific local experts in mixture-of-experts models.Oct 19, 2026
• 24Code2Video-Compass: A Comprehensive Benchmark for General-Purpose Code-Based Video Generation across T2V, I2V, and V2VCode2Video-Compass is a benchmark evaluating multimodal language models on code-based video generation across text, image, and video inputs, revealing significant challenges in complex dynamic and state-dependent tasks.Oct 19, 2026
• 23Self-Distilling Future Video Representations for World Action ModelsFRED improves single-step future video representations in world action models through self-distillation and temporal conditioning, boosting robot policy performance on long-horizon and unseen tasks.Oct 20, 2026
• 18Heartbeat Thinking: Speculative Reasoning and Tool Execution While Listening in Realtime Voice AgentsA heartbeat-style speculative reasoning approach lets voice agents continue preparing tool calls during speech updates, improving coverage and reducing response latency.Oct 18, 2026
• 17Beyond Watching and Searching: Benchmarking Agents for Video-centered Deep Research ReportsResearchers introduce a benchmark and evaluation framework for video-centered deep research agents that synthesize external evidence into interactive reports linked to video moments.Oct 21, 2026
• 17RouteWeaver: Weaving Mode Selection and Execution into Unified LLM RoutingRouteWeaver unifies heterogeneous LLM routing modes via coordinated training and progressive exploration, improving accuracy across diverse reasoning tasks.Oct 18, 2026
• 15Is Complex Agentic Memory Always Needed? Naive RAG Is Stronger Than We ThoughtRaw retrieval-augmented generation over unprocessed conversation history outperforms complex agentic memory systems across multiple benchmarks, indicating that preserving raw evidence is more critical than elaborate memory transformations.Oct 21, 2026
• 15RLSplat: Resource-Linear Budget Control for Feed-forward 3D Gaussian SplattingRLSplat enables feed-forward 3D Gaussian splatting with linear reductions in latency, memory, and model size by allocating Gaussians regionally and decoding them via local attention.Oct 19, 2026
• 14Cola2: Co-Adaptive Hierarchical Language Modeling with Continuous Semantic Compression and Balanced Discrete GenerationCola2 improves hierarchical language generation by jointly optimizing continuous semantic compression, reconstruction-generation balance, and encoder-decoder co-adaptation via dynamic span segmentation, masked diffusion decoding, and...Oct 21, 2026
• 13FilmFlow: Agentic System for 3D Virtual Film CinematographyFilmFlow is an agentic framework that translates natural-language or video instructions into executable camera trajectories for 3D scenes using specialized director, interpreter, scout, and critic agents, evaluated on the CinemaBench...Oct 21, 2026
• 13Agent Mosaic: Recognizing Confidentiality Leaks in Agentic LLM SystemsLLM agents frequently leak sensitive conclusions assembled from multiple confidential documents, and common safeguards fail to eliminate this mosaic-style disclosure risk.Oct 20, 2026
• 12Rethinking Sequence Compression with Test-time RegressionTest-time regression provides a learnable, closed-form alternative to pooling for compressing sequences in efficient language and multimodal models.Oct 22, 2026
• 12Verus-Insight: Towards Explainable Artifact Detection in AI-Generated VideosVerus-Insight improves AI-generated video detection by training artifact detectors through structured inspection planning, evidence verification, and evidence aggregation with human-anchored supervision.Oct 21, 2026
• 12RTAformer: Relation-Transition-Aware Transformer for Video Scene Graph GenerationRTAformer separately models persistent entities and changing relations in video scene graphs by injecting calibrated query differences and aligning cross-frame instances to overcome relational semantic inertia.Oct 13, 2026
• 11AnyGameRenderer: Entity-Consistent Generative Rendering Across Games and an Entity-Level BenchmarkAnyGameRenderer is a video-to-video generative model that transforms low-quality agent-built game footage into high-fidelity visuals while preserving layout, camera, and entity timing.Oct 21, 2026
• 11AgentVQA: A Unified Benchmark for Agentic Visual UnderstandingAgentVQA is an offline benchmark assessing vision-language models on realistic agentic tasks across web, robotics, video, games, and spatial planning, revealing that visual grounding and state tracking errors are the primary failure modes.Oct 21, 2026
• 8MolLean: A Process-Verifiable Language for Chemical ReasoningMolLean introduces a verifiable chemical reasoning framework with explicit molecular states, Lean-based verification, and interactive tool feedback to reduce hallucinations across generation and reaction tasks.Oct 13, 2026
• 7Inherited, Not Invented: Tracing the Provenance of Wait in Reasoning ModelsReinforcement learning improves reasoning accuracy without requiring a specific wait marker, which originates from training data rather than RL, and automated evaluation metrics can misrepresent reasoning trends.Oct 22, 2026
• 6RIDAR: Safeguarding Research Integrity in Autonomous Research Agents via Automated Detection and Runtime InterventionRIDAR detects and mitigates research integrity violations in autonomous agents via taxonomy-based post-hoc analysis and real-time intervention.Oct 21, 2026
• 6SCOPE: Compiling Natural-Language Policies into Neurosymbolic Predicate ProgramsSCOPE compiles compound policies into optimized neural-symbolic programs that achieve high accuracy, stability, and interpretability on inexpensive models.Oct 21, 2026
• 6Reading, Writing, Sharing, and Interacting with the World through VideoThis survey unifies video understanding and generation through world modeling, organizing progress around inferring, synthesizing, sharing, and interacting with latent temporal states to emphasize persistent dynamics over clip-level...Oct 21, 2026
• 5Different Tasks, Different Timesteps: Temporal Task-Aware Quantization for Diffusion Language ModelsT2Quant improves diffusion language model quantization by dynamically allocating activation precision across denoising steps based on predicted task- and time-dependent sensitivity.Oct 21, 2026
• 5What Makes a Good Unmasking Policy? A Controlled Study of Learned Unmasking in Diffusion Language ModelsOptimized unmasking policies for masked diffusion language models improve efficiency by using trajectory-contraction targets with confidence and hidden states, achieving better speed-accuracy trade-offs with minimal overhead.Oct 21, 2026
• 5F2G-Pose: Geometry-Aware Foundation Feature Lifting for Direct RGB-D Category-Level Object Pose EstimationF2G-Pose is a real-time feed-forward framework that predicts category-level object pose, metric size, and completed shape from a single RGB-D image using dense visual features, DGCNN tokenization, and a Mixture-of-Experts Transformer.Oct 21, 2026
• 4Foundation4D: Towards a unified perception and reconstruction model for autonomous vehiclesFoundation4D is a unified feed-forward model that jointly estimates geometry, detects objects, predicts semantic occupancy, and reconstructs Gaussians from multi-camera driving video, achieving state-of-the-art results across several...Oct 20, 2026
• 4Who Takes the Lead? Model Ability and Task-Conditioned Collaboration with HumansResearchers use a shared-control maze to show that stronger language-model agents selectively involve humans based on route advantage rather than failing to collaborate.Oct 18, 2026
• 3BrickWright: Agentic Creation of Physically Feasible Brick BuildsBrickWright improves physical validity of AI-generated LEGO assemblies by providing agents with an assembly API and feasibility validator, achieving perfect validity while preserving prompt adherence.Oct 21, 2026
• 3MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action ModelsMambaAct replaces the transformer action expert in flow-matching VLA policies with a compact Mamba-based expert and a context compressor to cut inference latency while improving robotic manipulation success.Oct 15, 2026
• 2Skeleton-Grounded Latent Action Model for Robot World ModelsSkeLAM decomposes latent actions into skeleton-guided kinematic and residual components to improve controllable robot world modeling and motion transfer.Oct 21, 2026
• 2SplatLab: Autonomous Scientific Discovery for 3DGS ResearchSplatLab autonomously reproduces 3D Gaussian Splatting baselines, proposes extensions, and writes manuscripts with fully traceable training logs, improving PSNR by up to 3.78 dB.Oct 20, 2026

Upvotes from Hugging Face Papers. Summaries as shown there.