Despite the remarkable progress of Vision-Language Models (VLMs) in adopting "Thinking-with-Images" capabilities, accurately evaluating the authenticity of their reasoning process remains a critical challenge. Existing benchmarks mainly rely on outcome-oriented accuracy, lacking the capability to assess whether models can accurately leverage fine-grained visual cues for multi-step reasoning. To address these limitations, we propose ViEBench, a process-verifiable benchmark designed to evaluate faithful visual reasoning. Comprising 200 multi-scenario high-resolution images with expert-annotated visual evidence, ViEBench uniquely categorizes tasks by difficulty into perception and reasoning dimensions, where reasoning tasks require utilizing localized visual details with prior knowledge. To establish comprehensive evaluation criteria, we introduce a dual-axis matrix that provides fine-grained metrics through four diagnostic quadrants, enabling transparent diagnosis of model behavior across varying task complexities. Our experiments yield several interesting observations: (1) VLMs can sometimes produce correct final answers despite grounding on irrelevant regions, and (2) they may successfully locate the correct evidence but still fail to utilize it to reach accurate conclusions. Our findings demonstrate that ViEBench can serve as a more explainable and practical benchmark for comprehensively evaluating the effectiveness of agentic VLMs.

Large language models (LLMs) have achieved remarkable success across many domains, yet their generative capabilities have been misused to produce highly realistic fake news that threatens social stability. However, the main existing detection methods, including small language model (SLM)-based approaches and LLM-based detectors, exhibit substantial performance degradation on LLM-generated content, reflecting a lack of robustness. SLM-based detectors rely heavily on fixed data distributions and shallow textual cues, while LLM-based detectors depend on prompt-based inference, making them sensitive to prompting and vulnerable to LLM hallucinations. To address this challenge, we propose Fake News Court (FNC), a robust multi-agent adversarial framework that integrates the complementary strengths of LLM and SLM for detecting LLM-generated fake news. FNC adopts an adversarial learning paradigm in which a quality-controlled generator produces diverse and challenging fake news, while a hybrid detector combines multi-dimensional LLM-based agent reasoning with an SLM-based classifier to ensure stable decisions. Extensive experiments show that FNC achieves an overall accuracy improvement of approximately 6% over competitive baselines on LLM-generated fake news, confirming its robustness.

Vision-Language Models (VLMs) excel at many multimodal tasks, yet they frequently struggle with tasks requiring precise understanding and handling of fine-grained visual elements. This is mainly due to information loss during image encoding or insufficient attention to critical regions. Recent work has shown promise by incorporating pixel-level visual information into the reasoning process, enabling VLMs to access high-resolution visual details during their thought process. However, this pixel-level information is often overused, leading to inefficiency and distraction from irrelevant visual details. To address these challenges, we propose the first framework for adaptive pixel reasoning that dynamically determines necessary pixel-level operations based on the input query. Specifically, we first apply operation-aware supervised fine-tuning to establish baseline competence in textual reasoning and visual operations, then design a novel rollout-guided reinforcement learning framework relying on feedback of the model’s own responses, which enables the VLM to determine when pixel operations should be invoked based on query difficulty. Experiments on extensive multimodal reasoning benchmarks show that our model achieves superior performance while significantly reducing unnecessary visual operations. Impressively, our model achieves 73.4% accuracy on HR-Bench 4K while maintaining a tool usage ratio of only 20.1%, improving accuracy and simultaneously reducing tool usage by 66.5% compared to the previous methods.