Bohan Zeng

Our new work, OpenWorldLib, is a unified inference codebase for advanced world models. It aims to provide a clearer definition of world models and, within a unified framework, evaluate the upper bounds of existing world model-related methods. Code link: github.com/OpenDCAI/OpenW…

--Towards Scalable Sidewalk Autonomy (2/4)-- Introducing UrbanVerse — turning city-tour videos into physics-aware, interactive simulations for scalable robot learning in NVIDIA Omniverse. 🌆 Real-world scene layouts 🧱100K+ 3D assets ICLR’26 paper and **Open-source** assets at 🔗 urbanverseproject.github.io We thank @CocoRobotics for robot support and NVIDIA Academic Grant support @NVIDIARobotics

Genie3 generates videos. We generate 𝟯𝗗 𝘄𝗼𝗿𝗹𝗱𝘀 you can actually use. Launching tomorrow — Tencent #HYWorld 2.0, an engine-ready World Model🚀 This isn't a video. It's a real 3D scene, all generated & editable. One image in. A whole 3D world out. 🔥Open-source tomorrow

@jbniu25 Excellent work! congratulations

MinerU2.5-Pro is out. Instead of chasing new architectures, we focused on data: scaling from <10M to 65.5M pages with better sampling & annotation. Same 1.2B model, but now #1 on OmniDocBench v1.6—beating both OCR specialists and large general models.

🌧️❄️🌨️Procedural rain and snow as Gaussian splats — infinite particles that move with the camera, fully rendered on the GPU. Add fog and it starts to feel like actual weather @playcanvas

@FishCao2059 @yangyang200413 别骂了别骂了

@yangyang200413 最可怜的那批人是没有ai但是有翟天临的人

不敢想象没有ai的时代前辈是怎么徒手搓出毕业论文的😄

repo link: github.com/OpenDCAI/OpenW…

Thanks for your interest! We're building a unified calling standard — this work serves as a template for our next steps. We'd love to hear the community's views on world models. If you have different opinions or work to promote, feel free to open an issue in our GitHub repo~

@Julien4Future @_akhaliq Thank you very much. We have built a standardized community for now, and we will try to improve the calling speed in the future, hoping to contribute to the community.

@_akhaliq Finally someone building world models as infrastructure, not research demos. Mobile-first implications for on-device AI are massive.

OpenWorldLib A Unified Codebase and Definition of Advanced World Models paper: huggingface.co/papers/2604.04…

@techpupparent @_akhaliq Thanks for your attention. That's a critical issue. We considered inter-block communication before release, but due to instability and varying dependencies of most current methods, this repo currently lacks communication between synthesis and reasoning for most pipelines.

@_akhaliq Good catch. The persistence/real-time split is elegant, but retrieval prioritization across multimodal signals without a mechanism? That's where the abstraction leaks.

@jatingargiitk @_akhaliq Thank you for your interest. Our repository aims to establish unified calling standards. Since world models are still in early development, the coupling between pipelines is relatively loose, and we are ready to iterate and update at any time. Thanks for your attention.

@_akhaliq curious if this actually becomes a usable codebase or just the usual “here are 9 model families under one repo” thing. unified definitions are nice, but the pain is always eval + reproducibility once people start swapping data, simulators, and toolchains.

@serrylei2020 Thank you!😁

@bohan_zeng A Nice work led by Bohan ! Introducing the precise concept of world model.

Thanks a lot for the shoutout! 🚀 If anyone is interested in a unified definition and calling standard for world models, feel free to check out our code and open an issue: github.com/OpenDCAI/OpenW… More training-optimized versions coming in our next project!

@sirshibaninja @HuggingPapers Thanks! Bias in perception & reasoning is indeed challenging. In this project, we focus on unifying SOTA methods and defining world model scope. We'll address bias in our next project with fine-tuning code, exploring better multimodal representations.

@HuggingPapers This looks like a solid framework for world models. How do you plan to handle potential biases in the perception and reasoning stages?

OpenWorldLib A unified codebase and standardized framework for advanced world models, integrating perception, interaction, and long-term memory capabilities to understand and predict complex environments across generation and reasoning tasks.

🚀A Simple Baseline for Streaming Video Understanding🚀 #SimpleStream reveals a simple baseline of *recent N-frame sliding window* beats SOTA memory-based methods on standard streaming video benchmarks @lmmslab - Project: simple-stream.github.io - Code: github.com/EvolvingLMMs-L…

DataFlex A Unified Framework for Data-Centric Dynamic Training of Large Language Models paper: huggingface.co/papers/2603.26…

DataFlex A unified data-centric training framework built on LLaMA-Factory, supporting dynamic sample selection, domain mixture adjustment, and sample reweighting with full DeepSpeed ZeRO-3 compatibility.

@Skywork_ai excellent work! we will add Matrix-Game-3 to github.com/OpenDCAI/OpenW… , as soon as possible!

Skywork Matrix-Game 3.0 is here! FULLY OPEN SOURCE! Real-Time and Streaming Interactive World Model with Long-Horizon Memory - Fully open source: code, model, and technical report - 720p @ 40FPS with a 5B model - Minute-long memory consistency - Trained on Unreal Engine + AAA games + real-world data - Scales up to 28B MoE for quality, dynamics, and generalization Homepage 👉 matrix-game-v3.github.io Code 👉 github.com/SkyworkAI/Matr… Model 👉 huggingface.co/Skywork/Matrix… Tech report 👉 github.com/SkyworkAI/Matr… Create. Explore. Play. With Matrix-Game 3.0

@Jiaxi_Cui world model的噱头感觉更早，不过是学生们搞研究的一个方向，顺带如果想宣传自己工作的同学可以来我们仓库提issue或者pr：github.com/OpenDCAI/OpenW…

冷知识，AI 视频已经发展三年了

我们预计下周会发布OpenWorldLib设计思路的技术报告 github.com/OpenDCAI/OpenW…，以及我们对于world model相关的思考，如果有相关工作需要宣传的研究者，欢迎在issue中提出，我们会在report中进行引用，这样也是对您方法的一个宣传～😄

@MersonVoice @Jiaxi_Cui 互联网不是...算了，哥你喜欢就好😊

@Jiaxi_Cui 我是M 这怎么说🤓

如果你能在今天在 X 以中文起号成功，那这个世界上没有什么平台是你不能做的 因为你需要做到对 X 的三无小号的无脑言论，心如止水的水平。 之后再去小红书、抖音之类的平台，因为有审核和敏感词机制的保护，这些平台黑粉的言论在你看来甚至会有点可爱