Ovis2 is the latest version of the Ovis series models proposed by Alibaba's international team. Compared to the previous version 1.6, Ovis2 has significant improvements in data construction and training methods. It not only enhances the capacity density of small models but also greatly improves chain of thought (CoT) reasoning capabilities through instruction fine-tuning and preference learning. Additionally, Ovis2 introduces video and multi-image processing capabilities, and enhances multilingual abilities and OCR capabilities in complex scenarios, significantly increasing the model's practicality.

According to reports, the core of Tencent Yuanbao's recent upgrade lies in the application of the "Hunyuan + DeepSeek" dual-mode aggregation technology. In the past, DeepSeek mainly played the role of information extraction, similar to a "scanner." However, with the support of Tencent's Hunyuan multimodal technology, DeepSeek can now truly understand the details, atmosphere, and even hidden meanings within images. This transformation enables Tencent Yuanbao to "comprehend" images and provide its own analysis and interpretation.

Recently, the successful launch of the VLM-R1 project has brought new hope to this field. This project represents the successful migration of the R1 Method from the DeepSeek team into visual language models, indicating that AI's understanding of visual content will enter a whole new phase. The inspiration for VLM-R1 comes from last year's open-source R1 Method by DeepSeek, which leverages GRPO (Generative Reward Processing Optimization) reinforcement.

Recently, the Microsoft Research team, in collaboration with researchers from several universities, released a multimodal AI model called 'Magma'. This model is designed to process and integrate various types of data, including images, text, and videos, to perform complex tasks in both digital and physical environments. As technology continues to advance, multimodal AI agents are being widely applied in fields such as robotics, virtual assistants, and user interface automation. Previous AI systems typically focused on visual-language understanding or robotic operations, making it difficult to combine the two.

On February 19, Beijing Zhipu Huazhang Technology Co., Ltd. announced that its multimodal creation platform 'Qingying' has formed a strategic partnership with the AI character creation platform 'NianTa'. Together, they will explore a complete path from AI character design to animation creation, and promote the technological realization of IP consistency. This collaboration aims to deeply explore the needs of secondary dimension users and AI application scenarios, providing users with higher quality generation effects and faster inference speeds.

Today, Step-Chronicles and Geely Automobile Group announced the joint open-source of two models from the Step series of multimodal large models: the Step-Video-T2V video generation model and the Step-Audio voice model. Among them, the Step-Video-T2V video generation model is globally ahead in both parameter count and performance. This model has 30 billion parameters and can directly generate high-quality videos at 540P resolution with 204 frames, ensuring high information density and strong consistency in the generated content. Evaluation results show that

Recently, Baidu's Ernie Bot officially announced that starting from 12:00 AM on April 1, 2024, its services will be fully available for free to all PC and app users. This means that users will not need to pay any fees to experience the latest features of the Ernie Bot series models.

It is reported that with the continuous iteration and cost optimization of the Ernie Bot large model, Ernie Bot will be completely free starting from midnight on April 1st. At that time, users will be able to experience the latest models in the Ernie Bot series without any barriers, whether on PC or APP. This initiative will undoubtedly further enhance the user coverage and market influence of Ernie Bot.

On February 11, 2025, Beijing Zhipu Tech announced that its developed Agentic GLM has fully launched on Samsung's latest Galaxy S25 series smartphones. This collaboration marks the official market release of the world's first smartphone that supports multimodal AI interactions, providing users with an unprecedented intelligent experience. Zhipu Tech began research on Agentic intelligence agents in April 2023 and launched its first Agent product, AutoGLM, in October 2024, being the first to propose Phone.

Recently, Alibaba has welcomed a heavyweight figure in the field of AI. According to industry insiders, a top global artificial intelligence scientist has officially joined Alibaba and will focus on the research and application of foundational large models for AI To C business in the future. This scientist has over 20 years of experience in both industry and academia, with significant achievements in the field of multimodal AI, having led the publication of over a hundred top papers on large models.

According to 'Intelligent Emergence', former Midjourney Asia Vice President Wang Yueting has joined Zhipu, primarily responsible for multimodal products and marketing. Meanwhile, Zhipu's Chief Strategy Officer Zhang Kuo and Vice President Qu Teng recently left the company. As of the time of publication, Zhipu has not responded to this development. The new head of multimodal products and marketing, Wang Yueting, previously served as Vice President of Midjourney, a prominent AIGC company, overseeing operations in China. Her joining Zhipu is seen as a significant strategic adjustment for the company in terms of products and marketing.

The Dark Side of the Moon Company proudly launches its brand new SOTA (state-of-the-art) model—k1.5 Multimodal Thinking Model, marking a significant breakthrough in multimodal reasoning and general reasoning fields. This model not only possesses outstanding multimodal processing capabilities but also demonstrates exceptional general reasoning abilities, effectively addressing various complex tasks. The standout feature of the k1.5 model is its multimodal reasoning capability. It can process information from different modalities including text, images, and sounds simultaneously, providing more comprehensive and accurate results.

Recently, Beijing Dark Side of the Moon Technology Co., Ltd. announced a significant technological upgrade for its intelligent assistant Kimi, introducing the brand new k1.5 Multimodal Thinking Model. This model has reached industry-leading levels in multimodal reasoning and general reasoning capabilities, marking another breakthrough for Kimi in the field of artificial intelligence.

Recently, Zhejiang University and Alibaba DAMO Academy jointly released a remarkable study aimed at creating high-quality multimodal textbooks from educational videos. This innovative research not only provides new ideas for training large-scale language models (VLMs) but may also change the way educational resources are utilized. With the rapid development of artificial intelligence technology, the pre-training corpus of VLMs mainly relies on visual-text pairs and visually intertwined data. However, much of this current data comes from the web, where the correlation between text and images is weak, and the knowledge density is relatively low.

On January 15, 2025, Beijing Dark Side of the Moon Technology Co., Ltd. announced the official release of the new multimodal image understanding model moonshot-v1-vision-preview. This model enhances the multimodal capabilities of the moonshot-v1 model series, helping Kimi better understand the world. The Vision model boasts powerful image recognition capabilities, accurately identifying complex details and subtle differences in images, distinguishing similar yet different objects, whether they are food or animals.