Ant Group's Ling-3.2-flash upgrades native vision: The 560-billion parameter flagship model can process images to operate Blender and create 3D content.

Beating AI News (Dongcha) reports that Ant Group’s Lingbai has launched Ling-3.2-flash, bringing native multimodal capabilities to its 560-billion-parameter flagship model. The new model supports image, text, and video inputs, with a focus on enhancing complex visual understanding, 3D interaction, and multimodal agent capabilities. It retains the same total parameters (560 billion), activated parameters (25 billion), and 1 million-token context window as its predecessor Ling-3.1-flash, while expanding beyond text reasoning to native visual comprehension. The prior Ling-3.0-flash-VL already supported image and video inputs, but only had 1240 billion total parameters, 55 billion activated parameters, and a 256,000-token context window. In official demos, the new model can generate webpages from images, operate Blender to build 3D scenes, and read research paper PDFs containing formulas and charts, converting their algorithms into runnable code. Ant Group’s internal tests show the new model outperforms Claude Opus 5 (max) in three benchmarks: document understanding, complex visual reasoning, and image-to-webpage generation. Its Vision2Web score reaches 85.86, compared to Opus 5’s 80.20. On third-party platform DesignArena’s image-to-frontend generation leaderboard, Ling-3.2-flash ranks 10th with an Elo score of 1238. The model is now available for web-based experience and API access, with free access for the first two weeks.