ERNIE 5.0 Technical Report

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

Beyond Unimodal Shortcuts: MLLMs as Cross-Modal Reasoners for Grounded Named Entity Recognition

Grounded Multimodal Named Entity Recognition (GMNER) aims to extract text-based entities, assign them semantic categories, and ground them to corresponding visual regions. In this work, we explore the potential of Multimodal Large Language Models (MLLMs) to perform GMNER in an end-to-end manner, moving beyond their typical role as auxiliary tools within cascaded pipelines. Crucially, our investigation reveals a fundamental challenge: MLLMs exhibit $\textbf{modality bias}$, including visual bias and textual bias, which stems from their tendency to take unimodal shortcuts rather than rigorous cross-modal verification. To address this, we propose Modality-aware Consistency Reasoning ($\textbf{MCR}$), which enforces structured cross-modal reasoning through Multi-style Reasoning Schema Injection (MRSI) and Constraint-guided Verifiable Optimization (CVO). MRSI transforms abstract constraints into executable reasoning chains, while CVO empowers the model to dynamically align its reasoning trajectories with Group Relative Policy Optimization (GRPO). Experiments on GMNER and visual grounding tasks demonstrate that MCR effectively mitigates modality bias and achieves superior performance compared to existing baselines.

Beyond Rigid: Benchmarking Non-Rigid Video Editing

Despite the remarkable progress in text-driven video editing, generating coherent non-rigid deformations remains a critical challenge, often plagued by physical distortion and temporal flicker. To bridge this gap, we propose NRVBench, the first dedicated and comprehensive benchmark designed to evaluate non-rigid video editing. First, we curate a high-quality dataset consisting of 180 non-rigid motion videos from six physics-based categories, equipped with 2,340 fine-grained task instructions and 360 multiple-choice questions. Second, we propose NRVE-Acc, a novel evaluation metric based on Vision-Language Models that can rigorously assess physical compliance, temporal consistency, and instruction alignment, overcoming the limitations of general metrics in capturing complex dynamics. Third, we introduce a training-free baseline, VM-Edit, which utilizes a dual-region denoising mechanism to achieve structure-aware control, balancing structural preservation and dynamic deformation. Extensive experiments demonstrate that while current methods have shortcomings in maintaining physical plausibility, our method achieves excellent performance across both standard and proposed metrics. We believe the benchmark could serve as a standard testing platform for advancing physics-aware video editing.

BayesRAG: Probabilistic Mutual Evidence Corroboration for Multimodal Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has become a pivotal paradigm for Large Language Models (LLMs), yet current approaches struggle with visually rich documents by treating text and images as isolated retrieval targets. Existing methods relying solely on cosine similarity often fail to capture the semantic reinforcement provided by cross-modal alignment and layout-induced coherence. To address these limitations, we propose BayesRAG, a novel multimodal retrieval framework grounded in Bayesian inference and Dempster-Shafer evidence theory. Unlike traditional approaches that rank candidates strictly by similarity, BayesRAG models the intrinsic consistency of retrieved candidates across modalities as probabilistic evidence to refine retrieval confidence. Specifically, our method computes the posterior association probability for combinations of multimodal retrieval results, prioritizing text-image pairs that mutually corroborate each other in terms of both semantics and layout. Extensive experiments demonstrate that BayesRAG significantly outperforms state-of-the-art (SOTA) methods on challenging multimodal benchmarks. This study establishes a new paradigm for multimodal retrieval fusion that effectively resolves the isolation of heterogeneous modalities through an evidence fusion mechanism and enhances the robustness of retrieval outcomes. Our code is available at https://github.com/TioeAre/BayesRAG.

Measuring Social Bias in Vision-Language Models with Face-Only Counterfactuals from Real Photos

Vision-Language Models (VLMs) are increasingly deployed in socially consequential settings, raising concerns about social bias driven by demographic cues. A central challenge in measuring such social bias is attribution under visual confounding: real-world images entangle race and gender with correlated factors such as background and clothing, obscuring attribution. We propose a \textbf{face-only counterfactual evaluation paradigm} that isolates demographic effects while preserving real-image realism. Starting from real photographs, we generate counterfactual variants by editing only facial attributes related to race and gender, keeping all other visual factors fixed. Based on this paradigm, we construct \textbf{FOCUS}, a dataset of 480 scene-matched counterfactual images across six occupations and ten demographic groups, and propose \textbf{REFLECT}, a benchmark comprising three decision-oriented tasks: two-alternative forced choice, multiple-choice socioeconomic inference, and numeric salary recommendation. Experiments on five state-of-the-art VLMs reveal that demographic disparities persist under strict visual control and vary substantially across task formulations. These findings underscore the necessity of controlled, counterfactual audits and highlight task design as a critical factor in evaluating social bias in multimodal models.

When Abundance Conceals Weakness: Knowledge Conflict in Multilingual Models

Large Language Models (LLMs) encode vast world knowledge across multiple languages, yet their internal beliefs are often unevenly distributed across linguistic spaces. When external evidence contradicts these language-dependent memories, models encounter \emph{cross-lingual knowledge conflict}, a phenomenon largely unexplored beyond English-centric settings. We introduce \textbf{CLEAR}, a \textbf{C}ross-\textbf{L}ingual knowl\textbf{E}dge conflict ev\textbf{A}luation f\textbf{R}amework that systematically examines how multilingual LLMs reconcile conflicting internal beliefs and multilingual external evidence. CLEAR decomposes conflict resolution into four progressive scenarios, from multilingual parametric elicitation to competitive multi-source cross-lingual induction, and systematically evaluates model behavior across two complementary QA benchmarks with distinct task characteristics. We construct multilingual versions of ConflictQA and ConflictingQA covering 10 typologically diverse languages and evaluate six representative LLMs. Our experiments reveal a task-dependent decision dichotomy. In reasoning-intensive tasks, conflict resolution is dominated by language resource abundance, with high-resource languages exerting stronger persuasive power. In contrast, for entity-centric factual conflicts, linguistic affinity, not resource scale, becomes decisive, allowing low-resource but linguistically aligned languages to outperform distant high-resource ones.

IDRBench: Interactive Deep Research Benchmark

Deep research agents powered by Large Language Models (LLMs) can perform multi-step reasoning, web exploration, and long-form report generation. However, most existing systems operate in an autonomous manner, assuming fully specified user intent and evaluating only final outputs. In practice, research goals are often underspecified and evolve during exploration, making sustained interaction essential for robust alignment. Despite its importance, interaction remains largely invisible to existing deep research benchmarks, which neither model dynamic user feedback nor quantify its costs. We introduce IDRBench, the first benchmark for systematically evaluating interactive deep research. IDRBench combines a modular multi-agent research framework with on-demand interaction, a scalable reference-grounded user simulator, and an interaction-aware evaluation suite that jointly measures interaction benefits (quality and alignment) and costs (turns and tokens). Experiments across seven state-of-the-art LLMs show that interaction consistently improves research quality and robustness, often outweighing differences in model capacity, while revealing substantial trade-offs in interaction efficiency.

MotionAdapter: Video Motion Transfer via Content-Aware Attention Customization

Recent advances in diffusion-based text-to-video models, particularly those built on the diffusion transformer architecture, have achieved remarkable progress in generating high-quality and temporally coherent videos. However, transferring complex motions between videos remains challenging. In this work, we present MotionAdapter, a content-aware motion transfer framework that enables robust and semantically aligned motion transfer within DiT-based T2V models. Our key insight is that effective motion transfer requires \romannumeral1) explicit disentanglement of motion from appearance and \romannumeral 2) adaptive customization of motion to target content. MotionAdapter first isolates motion by analyzing cross-frame attention within 3D full-attention modules to extract attention-derived motion fields. To bridge the semantic gap between reference and target videos, we further introduce a DINO-guided motion customization module that rearranges and refines motion fields based on content correspondences. The customized motion field is then used to guide the DiT denoising process, ensuring that the synthesized video inherits the reference motion while preserving target appearance and semantics. Extensive experiments demonstrate that MotionAdapter outperforms state-of-the-art methods in both qualitative and quantitative evaluations. Moreover, MotionAdapter naturally supports complex motion transfer and motion editing tasks such as zooming.

MIAR: Modality Interaction and Alignment Representation Fuison for Multimodal Emotion

Multimodal Emotion Recognition (MER) aims to perceive human emotions through three modes: language, vision, and audio. Previous methods primarily focused on modal fusion without adequately addressing significant distributional differences among modalities or considering their varying contributions to the task. They also lacked robust generalization capabilities across diverse textual model features, thus limiting performance in multimodal scenarios. Therefore, we propose a novel approach called Modality Interaction and Alignment Representation (MIAR). This network integrates contextual features across different modalities using a feature interaction to generate feature tokens to represent global representations of this modality extracting information from other modalities. These four tokens represent global representations of how each modality extracts information from others. MIAR aligns different modalities using contrastive learning and normalization strategies. We conduct experiments on two benchmarks: CMU-MOSI and CMU-MOSEI datasets, experimental results demonstrate the MIAR outperforms state-of-the-art MER methods.

Multimodal Sentiment Analysis based on Multi-channel and Symmetric Mutual Promotion Feature Fusion

Multimodal sentiment analysis is a key technology in the fields of human-computer interaction and affective computing. Accurately recognizing human emotional states is crucial for facilitating smooth communication between humans and machines. Despite some progress in multimodal sentiment analysis research, numerous challenges remain. The first challenge is the limited and insufficiently rich features extracted from single modality data. Secondly, most studies focus only on the consistency of inter-modal feature information, neglecting the differences between features, resulting in inadequate feature information fusion. In this paper, we first extract multi-channel features to obtain more comprehensive feature information. We employ dual-channel features in both the visual and auditory modalities to enhance intra-modal feature representation. Secondly, we propose a symmetric mutual promotion (SMP) inter-modal feature fusion method. This method combines symmetric cross-modal attention mechanisms and self-attention mechanisms, where the cross-modal attention mechanism captures useful information from other modalities, and the self-attention mechanism models contextual information. This approach promotes the exchange of useful information between modalities, thereby strengthening inter-modal interactions. Furthermore, we integrate intra-modal features and inter-modal fused features, fully leveraging the complementarity of inter-modal feature information while considering feature information differences. Experiments conducted on two benchmark datasets demonstrate the effectiveness and superiority of our proposed method.