multi-modal

The paper introduces Multi3IR, a benchmark for multi-perspective, multi-domain, multi-modal information retrieval, and proposes SPIN, a method to improve perspective coverage in retrieval systems.

DICS introduces a self-scoring metric for visual instruction data selection that enhances vision-language model performance with significantly less data by ensuring intra-sample consistency, outperforming state-of-the-art methods.

Poly-InstructTTS is a text-to-speech system that learns expressive speech from open-ended natural language instructions using a large-scale multi-modal dataset, improving instruction adherence and expressiveness in TTS models.

The tweet announces that Ox Alpha, equivalent to Gemini 4.0, is available for free on OpenCode for a week, featuring 1M context, multi-modal capabilities, and zero data retention.

A stealth AI model named Ox-Alpha has been released, reportedly outperforming Fable on SWE benchmarks, and is available for free with features like multi-modal support and zero data retention.

OpenRouter has released a new stealth AI model named Ox Alpha, optimized for efficient coding and agentic tasks with a 1M token context window and support for text, image, and video inputs.

This paper diagnoses perceptual-decision misalignment in Omni-LLMs and proposes a training-free inference-time framework called Modality Subspace Activation to mitigate it by dynamically balancing modality strengths.

This paper proposes a Multi-Modal Generative Fuzzy System (MMGFS) to enhance multimodal question answering by addressing modality bias and uncertainty through fuzzy inference and multi-hop reasoning, demonstrating improved performance on multiple benchmarks.

This paper introduces MAG, a manifold-guided framework for semi-supervised multi-modal in-context demonstration selection, leveraging unlabeled data to improve few-shot ICL for MLLMs. Experiments on eight benchmarks show consistent gains in label-scarce regimes.

RedNote has released dots3-note preview, the first open-weight model in the dots3 family, featuring 280B parameters with 16B active, multi-modal understanding (text, image, video, audio), 512K context length, and Apache 2.0 license, with strong agent capabilities.

A survey paper systematically analyzing the evolving safety landscape of multi-modal large language models, covering emerging threats such as adversarial attacks, data poisoning, jailbreaks, and hallucinations, and reviewing updated safety strategies.

Introduces GEOID-Flood, a large-scale multi-modal benchmark dataset for flood segmentation with over 14,000 tiles from 219 events across 65 countries, evaluating foundation models against conventional encoders across single-image, multi-temporal, and multi-modal protocols.

VideoAgent is an all-in-one open-source framework for comprehensive video intelligence, combining understanding, editing, and creative generation through a unified agentic workflow.

mere.run is a local-first inference runtime for Apple Silicon and headless Linux that provides a single CLI for text, image, video, music, 3D, and more without requiring Python.

This paper proposes Expert-Guided Mutual Distillation (EGMD) to address domain bias and semantic misalignment in multimodal fake news detection, achieving state-of-the-art accuracy and reducing domain bias by up to 57.3% across four datasets.

This paper introduces Explorative Modeling, a new generative modeling paradigm that factors the training loop by exploring candidate matches between model generations and data. It establishes a third pretraining axis beyond parameters and data, improves scaling efficiency across images, video, and language, and enables end-to-end generative modeling with far fewer inference steps.

StatePlay proposes a state-aware game world model that jointly predicts visual content and game states to generate mechanics-consistent game rollouts, achieving 18.6% improvement in mechanics fidelity.

Introduces VlogReward, a reward model for evaluating vlog editing plans across six dimensions, along with a large-scale dataset and benchmark, achieving state-of-the-art results against GPT-5 and Gemini-3-Pro.

GLI-AL is a new label resource for glioma MRI that unifies anatomy and lesion labels, expanding supervision to include healthy tissues and previously unlabeled abnormalities. It provides 1,251 label sets aligned with BraTS-GLI cases.

The paper proposes a Regime-Aware Multi-Modal Learning (RAML) method for Bitcoin price direction prediction that adaptively fuses social sentiment and technical features based on market volatility. Evaluated on hourly data from July 2024 to September 2025, RAML achieves moderate improvements over static fusion baselines.