Kimi K2.5: Visual Agentic Intelligence

Kimi Team

• Kimi K2.5 is a frontier visual agentic model that achieves state-of-the-art performance on computer-use and multimodal reasoning benchmarks, advancing open-source visual agentic intelligence.

[Paper]

Towards Cross-View Point Correspondence in Vision-Language Models

Yipu Wang^*, Yuheng Ji^*, Yuyang Liu^*, Enshen Zhou, Ziqiang Yang, Yuxuan Tian, Ziheng Qin, Yue Liu, Huajie Tan, Cheng Chi, Zhiyuan Ma, Daniel Dajun Zeng, Xiaolong Zheng

• Cross-view correspondence is a fundamental capability for spatial understanding and embodied AI. We propose the Cross-View Point Correspondence (CVPC) task and CrossPoint-Bench, a comprehensive benchmark with hierarchical design. Our evaluation shows state-of-the-art models still fall far behind humans. We construct CrossPoint-378K, a dataset with 378K question-answering pairs across 900 scenes, and propose CroPond that achieves state-of-the-art performance on CrossPoint-Bench, surpassing Gemini-2.5-Pro by 39.7% accuracy.

[Paper] [Code]

VisualTrans: A Benchmark for Real-World Visual Transformation Reasoning

Yuheng Ji^*, Yipu Wang^*, Yuyang Liu, Xiaoshuai Hao, Yue Liu, Yuting Zhao, Huaihai Lyu, Xiaolong Zheng

• VisualTrans is the first real-world benchmark for Visual Transformation Reasoning (VTR), evaluating spatial, procedural and quantitative reasoning across 12 human-object interaction tasks. While current models perform well on static tasks, they show significant limitations in dynamic, multi-step reasoning, revealing critical gaps in temporal and causal understanding for intelligent systems.

[Paper] [Code]

Wavelet attention-powered neural network framework with hierarchical dynamic frequency learning for lithium-ion battery state of health prediction

Yipu Wang, Huan Wang

• We propose WAPHF, a wavelet attention-powered hierarchical dynamic frequency learning framework for lithium battery SOH prediction. By integrating CNN with wavelet transform and dynamic frequency-focused attention, our method effectively addresses frequency aliasing issues and outperforms state-of-the-art approaches across three datasets.

[Paper]

OpenCUA: Open Foundations for Computer-Use Agents

Xinyuan Wang^*, Bowen Wang^*, Dunjie Lu^*, Junlin Yang^*, Tianbao Xie^*, Junli Wang^*, et al., Yipu Wang, Heng Wang, Diyi Yang, Victor Zhong, Flood Sung, Y.Charles, Zhilin Yang, Tao Yu

• We present OpenCUA, a comprehensive open-source framework for scaling CUA data and foundation models which includes an annotation infrastructure, the first large-scale computer-use task dataset and a scalable pipeline that transforms demonstrations into state–action pairs with reflective long Chain-of-Thought reasoning. Our end-to-end agent model, OpenCUA-32B achieves an average success rate of 32.5% on OSWorld-Verified, establishing a new state-of-the-art (SOTA) among open-source models and surpassing OpenAI CUA (GPT-4o).

[Paper] [Code]

Scaling Up AI-Generated Image Detection with Generator-Aware Prototypes

Ziheng Qin, Yuheng Ji, Renshuai Tao, Yuxuan Tian, Yuyang Liu, Yipu Wang, Xiaolong Zheng

• We propose a generator-aware prototype framework for scalable AI-generated image detection that leverages generator-specific knowledge to improve generalization across diverse generative models, achieving state-of-the-art performance on large-scale benchmarks.

[Paper]

Spatial Intelligence from a Cognitive Map Perspective: A Survey

Yuxuan Tian, Yuheng Ji, Xiaolong Zheng, Ziheng Qin, Yipu Wang, Xinyi Zheng, Yuyang Liu, Shuanghao Bai, Zhe Li, Liang Wang, Daniel Dajun Zeng

• A comprehensive survey on spatial intelligence in AI systems from a cognitive map perspective, covering spatial reasoning, scene understanding, and embodied AI with systematic categorization of current methods and future directions.