Global AI News Summary – January 28, 2026

1. DeepSeek Launches New OCR Model: AI Now Understands Images Like Humans

DeepSeek has officially introduced its new OCR model, DeepSeek-OCR 2, marking a significant advancement in text recognition technology. This model moves away from traditional OCR methods that rely on fixed scanning sequences (such as left-to-right or top-to-bottom) and instead employs the self-developed DeepEncoder V2 architecture. This architecture enables the model to deeply perceive the overall structure and semantic relationships of images, allowing it to identify key areas, assess the logic between text and images, and dynamically plan reading paths. For instance, when facing complex documents with tables, illustrations, multi-column text, and handwritten annotations, the model can autonomously determine information prioritization and spatial relevance, significantly enhancing recognition accuracy and logical coherence.

2. Ant Group’s LingBot-Depth Model Released: Empowering Robots and Autonomous Vehicles with 3D Vision

On January 27, Ant Group’s cutting-edge technology company, Ant Lingbo, unveiled its first open-source spatial perception model, LingBot-Depth. This model focuses on high-precision, low-latency monocular depth estimation and 3D spatial understanding, with lightweight deployment capabilities suitable for mobile devices and edge hardware. On the same day, Ant Lingbo announced a strategic partnership with leading domestic 3D visual sensing company Orbbec. Together, they aim to accelerate the application of spatial intelligence technologies across various high-potential fields, including service robots, spatial computing, smart cockpits, and industrial inspection.

3. Jack Ma on AI in Rural Education: AI is Not Here to Replace Teachers

Recently, Jack Ma participated online in the “Laba Festival Appointment” event organized by the Jack Ma Rural Public Welfare Foundation, focusing on the deep connection between AI and education. Addressing concerns from rural teachers about being replaced by AI, he emphasized that AI is not a threat but an opportunity for education to return to its essence. The key lies not in whether to use AI, but in teaching children how to use it effectively. As AI takes over memory and computation, humanity’s irreplaceable abilities—curiosity, imagination, creativity, judgment, and collaboration—will become the core of education.

4. Yang Jun from Conch Group Visits Huawei: Exploring AI Models for the Building Materials Industry

On January 26, Yang Jun, Secretary of the Party Committee and Chairman of Conch Group, visited Huawei for in-depth research and strategic discussions. The two parties discussed the development and implementation of AI models specific to the building materials industry, the path for traditional manufacturing to upgrade to smart technologies, trends in green and low-carbon transitions, and strategies for international development. This exchange represents a significant step forward in collaboration between a major state-owned enterprise and a leading technology company in the field of industrial intelligence.

5. Kimi Smart Assistant Upgrade: K2.5 Version Released Silently

On January 27, the highly anticipated Kimi Smart Assistant saw a key iteration with the silent launch of version K2.5. This upgrade seamlessly replaces the underlying model from K2 to K2.5 without requiring users to click on updates, log in again, or switch manually. This design significantly lowers the usage barrier, allowing all users to immediately experience enhanced understanding capabilities, improved long-text processing, and more natural multi-turn dialogue experiences.

6. Google AI Lead Orders Mac Mini: Clawdbot Becomes a 24/7 Digital Assistant for Everyone

In early 2026, an open-source AI assistant named Clawdbot became a viral sensation on GitHub, quickly garnering over 50,000 stars within days. Unlike traditional chatbots, it acts as a local, proactive AI agent capable of executing complex tasks such as ordering food, archiving emails, switching music, and making voice calls via messaging platforms like Telegram. Due to its lightweight deployment nature, the Mac mini has become an unexpectedly popular device for developers. However, the developers clarify that any device capable of running Node.js can be deployed, making it accessible to a broader audience.

7. SK Hynix Supplies Microsoft with Maia 200: Setting a New Benchmark in AI Chip Memory

SK Hynix has confirmed its role as the exclusive supplier of HBM3E high-bandwidth memory for Microsoft’s Maia 200 AI acceleration chip. This new chip is designed for training and deploying large models, featuring 216GB of ultra-large HBM3E memory, composed of six stacked 12-layer memory modules. Compared to the previous generation, HBM3E shows improvements in energy efficiency, signal integrity, and thermal design, making it better suited for the high concurrency data demands of data center-level AI chips.

8. Elon Musk Announces Starship V3 Launch in 6 Weeks: Upgrades Propel Human Moon Landing Plans

On January 26, Elon Musk announced via social media that SpaceX’s next-generation heavy-lift rocket, Starship V3, is set to make its first launch in six weeks. This version features significant upgrades in thermal protection, engine arrangement, and structural weight reduction. The upcoming test flight will focus on validating the rocket’s vertical recovery capabilities, super-heavy booster separation, and atmospheric re-entry stability, laying the groundwork for NASA’s Artemis moon landing plan and Mars transportation system.

9. OpenAI Launches Prism: A New AI-Native Workspace for Research

On January 27, OpenAI launched Prism, an AI-native workspace designed specifically for research purposes. Powered by the latest GPT-5.2 model, Prism focuses on core needs such as research writing, literature review, experiment documentation, and cross-team collaboration. This tool integrates into the research workflow, offering features like structured note-taking, automatic citation generation, collaborative editing, and intelligent content validation. All registered ChatGPT users can access it for free, with plans to gradually open it to organizations using the commercial and educational versions.

10. WeRide Unveils WeRide GENESIS: A Breakthrough in Physical and Generative AI Integration

On January 28, WeRide officially launched its self-developed general simulation model, WeRide GENESIS. This model combines the accuracy of physical modeling with the generalization capabilities of generative AI, allowing for high-fidelity reproduction of complex traffic flows, extreme weather, and real-world dynamics. By continuously feeding real vehicle data back into the simulation for optimization, GENESIS creates a feedback loop that significantly accelerates the algorithm iteration cycle and commercialization process in autonomous driving development.