On the 18th local time, Google officially launched the Gemini 3 series of artificial intelligence (AI) models and simultaneously launched the Gemini 3 Pro preview version. The company claims that Gemini 3 is the "most intelligent" and "most factual accurate" AI system to date. It not only has excellent reasoning ability, multimodal understanding ability, and intelligent agent coding ability, but also can quickly transform user creativity into applications. Compared with previous generation products, it has significantly improved in tasks such as creating software, organizing emails, and analyzing documents. Google also announced that starting from November 18th, all users can use its flagship model Gemini3Pro through the Gemini application, which will also be integrated into Google's search subscription service. According to Google, Gemini3's reasoning ability is top-notch, able to capture subtle clues in creativity, understand the deep meaning of complex problems, and more accurately grasp the context and intention behind user requests, thereby reducing reliance on a large number of prompt words. Some of its reasoning abilities have been demonstrated in independent benchmark tests. In the "Ultimate Human Exam" aimed at measuring general reasoning ability and professional knowledge, the model achieved its highest score ever with a score of 37.4. The previous highest score was maintained by GPT-5 Pro, at 31.64 points. Gemini3Pro has the ability of "native multimodality" and can simultaneously process text, images, and audio. For example, users can use it to translate recipe photos and generate electronic recipes, or convert a series of video lectures into interactive learning cards. In addition, Gemini3Pro also supports generating magazine style browseable pages or customized dynamic layout interfaces, providing a richer interactive experience. In search applications, Gemini3Pro can present diverse visual content such as images, tables, grids, and simulation results. The upgraded 'query decomposition technology' can not only break down problems into searchable subtasks, but also more accurately understand user intentions and help discover new content that may have been missed before. Google emphasizes that Gemini3Pro is different from OpenAI's ChatGPT in answering questions, as it is more "intelligent, concise, and direct", using real insights instead of clich é s, "telling you what you need to know, not just what you want to hear". Currently, Gemini3Pro has topped the popular AI benchmark platform LMArena rankings. Its' deep thinking 'mode further enhances reasoning ability, but this feature is currently only open to security testers. (New Society)