Every paper checked — nothing hidden
All papers
Every paper checked in the week of 12–18 Oct 2026 · Multimodal models — nothing is hidden. Brief papers first.
All1,924In brief3More (on-topic)553Not relevant1,333Not assessed19Older33Could not confirm19Blogs with no paper11All blogs13
| Paper | Topic | Relevance | Feedback |
|---|---|---|---|
| OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use The paper introduces an open-source framework and codebase for building generalist computer-use agents that operate across visual and code interfaces. In brief · #1New way to see a problemPreprint8/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Builds generalist computer-use agents combining visual interfaces and programmatic execution. | |
| VisualTrust: Video Reasoning under On-Screen Claim Framing Introduces a counterfactual benchmark to evaluate whether multimodal models erroneously defer to misleading on-screen text over video evidence. In brief · #2Surprising resultPreprint8/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Evaluates video-language models on text-video conflict and multimodal reasoning. | |
| Right Screen, Wrong Transition: World Models as Verifiers for GUI Agents The paper introduces LGWM, an action-conditioned world model predicting next-screen latent representations to verify GUI agent state transitions. In brief · #3Surprising resultPreprint8/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Evaluates visual state transitions for GUI agents using visual screen representations. | |
| Is In-Domain Training Enough for Fine-Grained Industrial Anomaly Understanding? Proposes a multi-agent framework combining general multimodal models with specialized visual experts for industrial defect detection and localization. Ranked #4Surprising resultPreprint8/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Investigates multimodal large language models performing industrial visual defect reasoning. | |
| Teaching Vision-Language Models to Decline: Support Completion Against Mirage The paper proposes a training method that teaches vision-language models to decline to answer when relevant visual evidence is missing. Ranked #5New way to see a problemPreprint8/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Trains vision-language models to decline ungrounded questions across invalid image inputs. | |
| Visually Legible but Adversarial: Watermark-Constrained Attacks on End-to-End VLMs for Document Understanding This paper introduces a question-agnostic adversarial text-watermarking method that severely degrades document understanding in vision-language models. Ranked #6Surprising resultPreprint8/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Analyzes adversarial watermark attacks on vision-language models for document understanding. | |
| Seeing Through the Forgery Is Not Enough: Regressive Sycophancy under Fabricated Evidence in Medical Vision-Language Models The authors show that medical vision-language models frequently abandon correct clinical answers when prompted with fabricated medical documents. Ranked #7Surprising resultPreprint8/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Studies multimodal medical vision-language models processing medical images and documents. | |
| MJev-Spatial: System One Spatial Intelligence via Native Attention Matching Frames spatial understanding tasks as candidate classification using pretrained attention matching to enable low-latency spatial decisions. Ranked #8New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Presents a multimodal model framework for fast spatial understanding in vision-language tasks. | |
| Any Time, Any Space: A World-State-Centric Framework for Agentic Video Generation This paper introduces Any Time, Any Space, a world-state memory framework enabling consistent multi-shot long-form agentic video generation. Ranked #9New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Long-form agentic video generation maintaining cross-shot spatiotemporal consistency. | |
| Player-1: Learning to Play from Synthetic Experience Develops a post-training framework that trains computer-use agents for video games using synthetic interactive experiences and an expanded action space. Ranked #10New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Trains multimodal models to perceive screen frames and execute mouse and keyboard actions. | |
| Speculative Forcing for Efficient Autoregressive Video Generation Accelerates autoregressive video generation by verifying and freezing intermediate tokens early to reduce redundant computation during denoising. Ranked #11Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Presents speculative early exit strategies for efficient autoregressive video generation. | |
| EviScene: Beyond "Is It AI?" for Agentic Image Factuality Assessment Introduces a benchmark and multimodal agent that uses visual inspection and external retrieval to verify real-world image factuality. Ranked #12New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Introduces a multimodal agent and benchmark combining visual inspection with external text retrieval. | |
| ExeWorldBench: Benchmarking Code-Based Executable 3D World Modeling The paper introduces a benchmark evaluating whether coding agents can construct interactive, code-based 3D simulations that capture underlying physical principles. Ranked #13New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Benchmarks translating visual observations into code-based executable 3D environments. | |
| Open-o3-Video-v2: Seek, Ground, and Reason with Verifiable Spatio-Temporal Evidence Develops an agentic video reasoning framework that iteratively identifies missing information, gathers visual evidence, and grounds temporal clues. Ranked #14New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Develops an agentic grounded reasoning framework enabling multimodal video models to retrieve evidence and verify reasoning. | |
| VIEWCRAFT: Evolving What Multimodal Agents See and How They Read It This paper proposes a framework that jointly evolves visual evidence presentation and reading procedures to improve multimodal agent task performance. Ranked #15New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Evolves visual observation interfaces and reading procedures for multimodal agents. | |
| Intern-Decision: Efficient Multimodal Decision Making with Autoregressive Masked Language Modeling Develops small multimodal models that output structured choices using masked language modeling to reduce inference latency in agent tasks. Ranked #16New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Evaluates models making structured decisions from text and visual image inputs. | |
| Elephant in the Fridge: Constant-Memory Frame Packing for Long Video Understanding Proposes a training-free frame packing method enabling vision-language models to process long videos with constant memory consumption. Ranked #17Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Compresses arbitrarily long video sequences into visual tokens for vision-language models. | |
| WOVEN: Weaving Visual World Modeling into Multimodal LLMs Presents a benchmark and dataset using generative video rollouts to train and evaluate visual transition reasoning in multimodal models. Ranked #18New testPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Explicitly targets multimodal LLMs to enhance spatial, embodied, and temporal visual reasoning. | |
| DecisionRobust: Benchmarking Robustness for JEV Style Decision Models Across Text and Vision The paper introduces a benchmark evaluating the robustness of typed decision models against text, image, and prompt injection attacks. Ranked #19New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Evaluates decision models across both textual and visual inputs under multimodal perturbations. | |
| Rethinking Projector Training in Multimodal LLMs The paper examines projector training in multimodal models and proposes a training-free initialization method using summary statistics instead of paired image-caption data. Ranked #20Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Investigates the alignment stage between vision encoders and language models in multimodal LLMs. | |
| What Really Works for Universal Multimodal Embeddings? A Controlled Study of Contrastive Training Strategies The paper conducts a controlled benchmark study evaluating common contrastive training strategies for universal multimodal embedding learning. Ranked #21Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Analyzes contrastive learning strategies for universal multimodal embeddings. | |
| SyncReward: Human-Aligned Synchronization Reward for Audio-Visual Generation The paper introduces a human-annotated dataset, benchmark, and reward model to improve temporal synchronization in audio-visual generation systems. Ranked #22New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Evaluates and aligns multimodal audio-visual generation and synchronization with human judgments. | |
| To Align or Not To Align: Check Your COMPASS Before You Train This paper introduces a metric to estimate the compatibility of pretrained models before running costly alignment or checkpoint merging. Ranked #23New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Provides a metric to evaluate compatibility when aligning vision encoders with language models. | |
| Auditing VLM Benchmarks from Existing Evaluation Runs Audits vision-language model benchmarks by analyzing saved evaluation artifacts and removing image inputs to detect ungrounded questions. Ranked #24New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Directly audits vision-language model benchmarks to identify multimodal image-text dependency failures. | |
| ConClick: Element-Wise Contrastive Learning Improves GUI Grounding Proposes a contrastive learning framework to help GUI agents distinguish target interface elements from plausible distractor elements. Ranked #25New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Enhances vision-language models for multimodal graphical user interface grounding. | |
| CV-OPD: Contrastive Vision On-policy Distillation Introduces Contrastive Vision On-policy Distillation to prevent vision-language models from relying on answer shortcuts rather than visual evidence. Ranked #26Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Addresses grounding failures and token constraints specifically in vision-language models (VLMs). | |
| Golden Anchor for Video Generation Studies how text-to-image anchor properties correlate with downstream video generation quality in image-to-video diffusion pipelines. Ranked #27Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Addresses text-to-image and image-to-video generation across visual and textual modalities. | |
| In-Context Learning Can Help Vision-Language Models Overcome Training Priors Demonstrates that visual in-context demonstrations help vision-language models overcome conflicting pretraining priors on counterfactual tasks. Ranked #28Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Studies overcoming training priors in multimodal vision-language models using visual in-context learning. | |
| Mind the Temporal Gap: Evaluating Video Models for Temporal Understanding in Driving Evaluates prompting methods and visual feature probing for temporal ego-motion reasoning in driving video-language models. Ranked #29Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Evaluates multimodal video-language models on temporal dynamic understanding tasks. | |
| What 30,000 Hours of Ego-centric Video Does Not Teach Studies how scaling to 30,000 hours of egocentric video affects world model fidelity for agents versus object interactions. Ranked #30Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Evaluates multimodal world models trained on large-scale ego-centric video for agent-object interaction. | |
| FearCaut-Qwen: Affective Steering in a Vision-Language Model Shifts the Decision Criterion for Hazard Assessment The paper uses activation steering of affective circuits in vision-language models to increase hazard-detection recall for post-disaster damage assessment. Ranked #31New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Studies internal representation steering in vision-language models for hazard assessment. | |
| Does Target Alignment Mean Target Recovery? An Evidence-Ladder Study of Adversarial Claims on Contrastive Encoders The paper evaluates whether contrastive encoder adversarial alignment metrics transfer to target recovery by independent models across different perturbation budgets. Ranked #32Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Investigates target alignment and adversarial transferability in contrastive vision-language encoders. | |
| When Charts Deceive Web Agents: A Paired Benchmark for Misleading Visualization-Grounded Tasks Introduces a benchmark evaluating how misleading visualizations and charts affect the multi-step execution and decisions of web agents. Ranked #33New testPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Tests vision-language models interpreting complex chart visual inputs to make decisions. | |
| DPOison-VL: Clean Preference Poisoning in Vision-Language Direct Preference Optimization The paper introduces an imperceptible image perturbation attack that manipulates alignment in vision-language models during preference optimization without altering text annotations. Ranked #34Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Investigates alignment and preference optimization specifically in vision-language models. | |
| Bifocal: Training-Free Mixed Granularity Visual Token Pruning for Multimodal LLMs Introduces Bifocal, a training-free mixed-granularity visual token pruning method for multimodal models to protect fine-grained task accuracy. Ranked #35Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Presents a visual token pruning method specifically designed to improve multimodal LLM efficiency and grounding. | |
| Connecting the Dots: Enduring Performance Gaps in ASR Systems Between Standard and Minority English Varieties The paper tracks automatic speech recognition performance from 2021 to 2025, finding persistent accuracy disparities between standard American English and minority accents. Ranked #36Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Evaluates speech-to-text multimodal audio processing across diverse English accents. | |
| Scaling Laws for Multimodal Data Mixtures The paper derives multimodal data mixing scaling laws across text, vision, and audio using mixture-of-experts models. Ranked #37Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Derives data mixture scaling laws across text, vision, and audio. | |
| Unsupervised Multimodal Reward Modeling Trains an unsupervised vision-language reward model using intermediate layer representations and caption perturbations without human preference labels. Ranked #38Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Develops an unsupervised reward model specifically for vision-language backbones and captions. | |
| Locate, Judge, Read: Role-Specialized Tool Agents over a Knowledge-Point Index for Open-Domain Multimodal Document QA Designs a modular multi-agent pipeline separating search, verification, and reading for multimodal document question answering. Ranked #39Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Handles reading and reasoning over multimodal documents combining text, images, and tables. | |
| When VLMs Destroy What They Know: Spatial Representations Fall Below Lexical Priors The paper evaluates vision-language models' spatial representations using causally controlled perturbations, showing spatial relations degrade compared to lexical features. Ranked #40Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Investigates how vision-language models fail to encode structured spatial relations internally. | |
| SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models Introduces a benchmark evaluating how well vision-language models predict unseen physical outcomes following scene interventions. Ranked #41New testPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Core contribution evaluates spatial reasoning benchmarks specifically designed for vision-language models. | |
| SteerablePlex: Can We Steer Full-Duplex Models? Introduces an instruction-following benchmark and training recipe to improve goal adherence in full-duplex conversational models. Ranked #42New testPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Applies to multimodal full-duplex speech and conversational audio models. | |
| When History Helps and Hurts: Selective History Use across Multimodal Turns The paper introduces a 7,000-task benchmark to evaluate how multimodal models selectively use or ignore conversational context. Ranked #43New testPreprint7/10 suggested | Multimodal models via arxiv · 12 Oct, 16:28 | On-topic Evaluates omni-modal, vision-language, and audio-language models across multimodal turns. | |
| FusionAudit: Pathway-Conditioned Robustness Auditing for Native Multimodal Models Develops a robustness auditing framework that evaluates multimodal models against pixel perturbations and competing rendered-text visual overlays. Ranked #44New testPreprint7/10 suggested | Multimodal models via arxiv · 17 Oct, 06:18 | On-topic Audits pathway-conditioned vulnerabilities and adversarial robustness in native vision-language models. | |
| Data-Free Continual Learning for World Action Model with Rectified Generative Replay This paper proposes a rectified generative replay framework enabling robotic world action models to continually learn new tasks without storing past data. Ranked #45Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Explores multimodal video generation models acting as world action models in robotics. | |
| Seg2Click: Learning Pixel-Level Action Masks for GUI Visual Grounding Reformulates graphical interface visual grounding as executable-region mask prediction to pinpoint click actions reliably. Ranked #46New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Investigates multimodal grounding of text instructions to visual regions in UI screenshots. | |
| Geometric Parrots: Rule Absence vs. Rule Fragility in Spatial and Physical Reasoning for Vision-Language Models Introduces a procedural benchmark and dual-loop evaluation to test whether vision-language models maintain physical and spatial reasoning rules under pressure. Ranked #47New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Benchmarks spatial, physical, and visual reasoning robustness in vision-language models. | |
| Language2Drive: Do Vision-Language-Action Models Really Follow the Instructions in Autonomous Driving? The paper introduces a counterfactual benchmark to evaluate whether vision-language-action driving models truly follow natural language instructions. Ranked #48Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Evaluates instruction following in multimodal vision-language-action models. | |
| Pixels or Code? Benchmarking Instruction-Guided Visual Artifact Operation The paper introduces a benchmark to evaluate direct edits, dependency updates, and visual preservation on technical diagrams, charts, and documents. Ranked #49New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Introduces a benchmark evaluating multimodal systems on instruction-guided editing of visual documents and diagrams. | |
| A Plausible Future Is Not a Causal Future: Can Video Models Generate How the World Unfolds? The paper presents a simulator-grounded benchmark to evaluate whether image-to-video generation models capture causal physical cascades and counterfactual dynamics faithfully. Ranked #50New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Evaluates causal physical dynamics and scene evolution in multimodal video generation models. | |
| RSAJEV: Training-Free Abstaining Prototype Memory for Eliminating Silent Hallucination in JEV-Based Remote Sensing Classification This paper introduces a training-free prototype memory readout to prevent out-of-vocabulary overconfidence and silent hallucinations in remote sensing classification. Ranked #51Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Investigates multimodal large language models for scene classification and hallucination reduction. | |
| Scenes as an Attack Surface: Uncovering Scene-Grounded Safety Failures in MLLMs The paper introduces a benchmark to show that visual scene semantics alone can cause multimodal models to bypass safety guardrails without adversarial prompts. Ranked #52Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Investigates scene-grounded visual safety failures and attack surfaces specifically in multimodal large language models. | |
| PVFH-Bench: Benchmarking Premise-Induced Visual Fixation Hallucinations in Vision-Language Models This paper introduces a benchmark to measure vision-language model hallucinations caused by deceptive textual premises paired with visually confusable images. Ranked #53Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Examines premise-induced object hallucination in large vision-language models. | |
| Meta-Video: Efficient Standardized Evaluation for Video Understanding Evaluates video benchmarks, showing that unexamined setup choices shift accuracy significantly and proposing a standardized evaluation framework. Ranked #54Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Introduces a standardized evaluation benchmark specifically for video-language multimodal models. | |
| Recovering Low-Bit Failures in Vision-Language-Action Policies via Quantization-Conditioned Reinforcement Learning This paper proposes offline reinforcement learning conditioned on quantization levels to recover task completion performance in quantized vision-language-action policies. Ranked #55Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:18 | On-topic Studies low-bit quantization recovery for multimodal vision-language-action models. | |
| Spurious Attention Heads in Vision Transformers: Mechanistic Identification of Annotation-Driven Bias in Medical Chest X-Ray Analysis Identifies a specific vision transformer attention head that spuriously attends to medical image corner text instead of lung pathology. Ranked #56Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 17 Oct, 06:17 | On-topic Studies Vision Transformers processing image data and text annotation corner features. | |
| Visual Dependence Does Not Imply Answerability: A Systematic Analysis of Answerability Signals in Frozen InstructBLIP Analyzes answerability signals in frozen visual question answering models, finding standard confidence and alignment metrics fail to reliably detect unanswerable inputs. Ranked #57Surprising resultPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Investigates answerability signals and visual grounding in multimodal vision-language models. | |
| VLA-GAUGE: Systematically Evaluating Generalization and Robustness in Vision-Language-Action Models Introduces a diagnostic evaluation framework to systematically test generalization and perturbation robustness in vision-language-action robotic models. Ranked #58New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 06:17 | On-topic Systematically evaluates multimodal Vision-Language-Action models across generalization tasks. | |
| SEA-HELM Audio: Evaluating Multilingual ASR-LLM Pipeline Performance Across SEA-HELM Subsets The paper presents an audio evaluation benchmark assessing accuracy drops between speech-input and text-input pipelines in Southeast Asian languages. Ranked #59New testPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Evaluates multimodal speech-and-text pipeline performance across multilingual audio inputs. | |
| Vision-Language Models Integrate Conflicting Evidence Like Biased Bayesian Observers The paper proposes a Bayesian cue-combination model to explain how vision-language models integrate conflicting visual and textual evidence. Ranked #60New way to see a problemPreprint7/10 suggested | Multimodal models via arxiv · 15 Oct, 15:06 | On-topic Investigates how vision-language models combine and resolve conflicting visual and textual evidence. |
1–60 of 1,924Next →