Alibaba Ovis-U1 Launches with a Bang: A Multi-Modal AI All-in-One, Open Source Empowers Global Developers

AIbase基地

Published in AI News · 6 minute read · Jul 9, 2025

On June 29, 2025, the Alibaba International AI team officially launched the new multimodal large model **Ovis-U1**, marking another major breakthrough in the field of multimodal artificial intelligence. As the latest masterpiece of the Ovis series, Ovis-U1 integrates multimodal understanding, image generation, and image editing capabilities, demonstrating powerful cross-modal processing abilities, and offering new possibilities for developers, researchers, and industry applications. The following is a detailed report from AIbase on Ovis-U1.

Ovis-U1: A Three-in-One Multimodal Unified Framework

Ovis-U1 is a 3-billion parameter model developed by the Alibaba International AI team based on the Ovis series architecture, which for the first time achieves unification of multimodal understanding, text-to-image generation, and image editing. According to AIbase, the model adopts an innovative architecture design, efficiently aligning visual and textual embeddings through three core components: a visual tokenizer, a visual embedding table, and a large language model (LLM). This structured alignment approach overcomes the limitations of traditional multimodal models in cross-modal transformation, significantly enhancing the model's performance in complex scenarios.

Ovis-U1 can process various input formats such as text and images, and it demonstrates excellent performance in tasks such as mathematical reasoning, object recognition, text extraction, and video understanding. For example, it can not only accurately identify objects or handwritten text in images but also generate high-quality images or perform precise edits on existing images according to user instructions. This "three-in-one" capability makes it highly promising for application in fields such as education, e-commerce, healthcare, and autonomous driving.

Technical Highlights: Efficient Training and Open-Source Sharing

The development of Ovis-U1 relies on advanced training strategies and diverse datasets. According to official information, the model is built using technologies such as Python 3.10, Torch 2.4.0, and Transformers 4.51.3. During the training process, DeepSpeed 0.15.4 optimization was adopted to ensure efficiency and stability. In addition, Ovis-U1 continues the open-source tradition of the Ovis series, using the Apache 2.0 license. The code, model weights, and training data are publicly available on Hugging Face and GitHub, allowing developers to quickly reproduce and deploy the model with simple environmental configuration.

AIbase noticed that Ovis-U1 introduced a compliance checking algorithm during training, ensuring that the model's outputs comply with ethical and legal requirements. This transparent development approach not only reflects Alibaba's contribution to the open-source community but also provides a convenient tool for global developers to explore multimodal AI.

Ovis-U1's multimodal capabilities make it perform well in practical applications. For instance, in the e-commerce sector, Ovis-U1 can analyze product images to generate multilingual descriptions or edit product display images according to user needs, enhancing consumer experience. In educational scenarios, it can recognize handwritten mathematical formulas and provide detailed solutions, assisting students in their learning. Additionally, Ovis-U1 supports generating recipes and analyzing video content, providing innovative solutions for smart homes and content creation.

AIbase believes that the release of Ovis-U1 not only solidifies Alibaba's leading position in the field of multimodal AI but also promotes the popularization and advancement of global AI technology through its open-source model. In the future, Ovis-U1 is expected to be applied in more industry scenarios, becoming an intelligent bridge connecting vision, language, and decision-making.

Since the release of Ovis-U1, there have been many discussions on social media. Many developers have praised the model's versatility and open-source features, considering it a low-barrier AI solution for small and medium-sized enterprises and individual developers. AIbase expects that with the widespread application of Ovis-U1, more innovative use cases will emerge within the community.

Project: (https://huggingface.co/AIDC-AI/Ovis-U1-3B)

Baidu Launches the WENXIN Large Model 4.5 Series Open Source, Sparking a New Wave in the Domestic Large Model Market!

Recently, Baidu officially announced the open-source release of its WENXIN Large Model 4.5 series, launching a total of ten models, including mixed expert (MoE) models with 47B and 3B activated parameters, as well as dense models with 0.3B parameters. This open-source initiative not only fully publicizes the pre-trained weights but also provides inference code, marking a significant advancement for Baidu in the field of large models. These newly released models can be downloaded and deployed on platforms such as PaddlePaddle Starry Sky Community and Hugging Face. Additionally, Baidu Intelligent Cloud's Qianfan Large Model Platform also provides

The Internal Testing Period of Xiaomi AI Toolbox Ends, Service Will Be Suspended Starting July 5

The internal testing project "Xiaomi AI Toolbox" has officially announced the end of its phased internal testing and plans to suspend service starting July 5, 2025. "AI Toolbox" is an important AI project incubated internally by Xiaomi, aimed at exploring and integrating cutting-edge AI technologies to provide users with a series of innovative features and experiences. Although the specific internal testing functions and application scenarios have not been fully disclosed, its name suggests its positioning as a multifunctional AI toolset. During the recent internal testing period, "AI Toolbox" has gathered some Xiaomi employees and core users.

The 'In-Depth Research' Feature of Doubao is Now in Testing on the Doubao APP, Web Version, and Desktop Version

Recently, the Doubao APP, web version, and desktop version platforms have introduced a new feature test - the 'In-Depth Research' feature has been officially launched, offering users free trial. This feature aims to help users efficiently handle complex tasks by quickly integrating massive in-depth information and generating detailed research reports or visualized web results.

Baidu Makes a Major Open-Source Release of the ERNIE Bot 4.5 Series with Ten New Models Unveiled!

Baidu officially released the ERNIE Bot 4.5 series models and made them fully open source. Users can experience this latest open-source technology immediately through ERNIE Bot (https://yiyan.baidu.com). This series includes multiple parameter configurations, such as Mixture of Experts (MoE) models with activated parameters of 47B and 3B, as well as dense models designed with 0.3B parameters, totaling ten different models. In terms of training and inference, the ERNIE 4.5 series models use PaddlePaddle deep learning.

Philips Launches the 8 Pro AI Earphones, Unlocking a New Multilingual Communication Experience!

At the 2025 new product launch event held recently in Xi'an, Philips Audio and Accessories Company introduced three groundbreaking products, attracting the attention of many tech enthusiasts. These new products include the "Blade" magnetic charging power bank designed for young white-collar workers and Generation Z, the 8 Pro AI Earphones tailored for business elites, and the KTV Cloud Speaker suitable for family entertainment. This series of innovative products not only highlights Philips' leadership in the consumer electronics field, but also marks the opening of a new market segment in China. Among them, the most eye-catching

Breaking News! GPT-5 is About to Arrive, Take You into a New Multimodal AI Era!

Recently, news about OpenAI's upcoming release of GPT-5 has attracted widespread attention in the technology industry. According to insiders, GPT-5 has already started a gradual test and is expected to be officially launched in July this year. This new model will adopt a multimodal design, meaning it can not only process text input but also understand speech, images, code, and even videos, completely changing the way we interact with AI. Sam Altman, CEO of OpenAI, stated that the launch of GPT-5 will mark a new era in AI.