Google officially launched "Gemini 3"! The official claim is that it is the "smartest AI model in the world". What are its features?

Google has officially announced the launch of its latest generation large language model “Gemini 3”. This is the third flagship AI model launched by Google in just two years, following Gemini 1 at the end of last year and Gemini 2 earlier this year, officially positioned as the “smartest Gemini ever.” (Background: Grok4's intelligence score beats OpenAI o3 and Gemini 2.5 Pro! Is the multi-faceted competition shifting to politics?) (Additional background: Is Gemini 3 coming? Frequent leaks on X suggest performance surpassing GPT-5 and Grok 4.) Tech giant Google released an announcement in Taiwan time on November 19, officially launching its latest generation large language model “Gemini 3,” which is positioned as the “smartest Gemini ever” following Gemini 1 and Gemini 2. Google executives stated: It has evolved from “understanding text” to “understanding context.” Google CEO Sundar Pichai stated in the announcement that the Gemini series has accumulated over 650 million monthly active users, with 2 billion instances using AI Overviews, and 13 million developers adopting it. He emphasized that Gemini 3 is not just an upgrade in parameters and speed, but represents a critical shift of AI moving from passive responses to “actively understanding intent and context.” At the same time, Google DeepMind CEO Demis Hassabis also stated: “Gemini 3 is currently the strongest model in the world for multimodal understanding and our most powerful agentic model.” Key advancements of Gemini 3 include: 1. Significant leap in reasoning ability, breaking through 1501 Elo on the LMArena leaderboard, ranking first. GPQA Diamond reached 91.9%, Humanity’s Last Exam 37.5% (without tools). The MathArena Apex for math problems set a new record of 23.4%. After launching the “Deep Think” mode, GPQA Diamond further increased to 93.8%. 2. Comprehensive leadership in multimodal understanding, with MMMU-Pro at 81% and Video-MMMU at 87.6%. It can process text, images, videos, audio, and code simultaneously, with a context length of up to 1 million tokens. 3. Comprehensive upgrade of agent and developer experience, with SWE-bench Verified at 76.2% and WebDev Arena at 1487 Elo. A new agent priority development platform “Google Antigravity” is now online, allowing AI agents to directly operate editors, terminals, and browsers. It supports real-time access to third-party tools such as Cursor, GitHub, and Replit. 4. Significant expansion of daily application scenarios. It can translate and organize family handwritten recipes into an electronic recipe book. Automatically generate training plans after analyzing sports competition videos. The “AI Mode” in search will now feature Gemini 3, providing real-time interactive visualization and simulation tools. Gradual rollout starts immediately. General users: Gemini App and AI Mode in search are immediately available (Pro/Ultra subscribers prioritized). Developers: Google AI Studio, Vertex AI, and Google Antigravity have opened the Gemini 3 Pro preview version. Enterprise users: Vertex AI and Gemini Enterprise are online simultaneously. After the safety evaluation of the “Gemini 3 Deep Think” mode, it will be available to Google AI Ultra subscribers within weeks. Google emphasizes that Gemini 3 has completed the strictest safety tests in history and has collaborated with institutions such as the UK AISI for third-party evaluations. This release is seen as Google's last major AI update before the end of 2025, entering a new round of intense competition with top models from OpenAI, Anthropic, xAI, and others. Related reports: The EU reportedly imposed a €2.95 billion “antitrust fine” on Google, marking the fourth fine in the past decade. Former Google CEO: Google is lagging in the AI race because it has been too comfortable, praising TSMC's competitive strength… the community is in an uproar. Google's strongest AI “Gemini” has been accused of fabrication! Google admits that the video was edited, not real-time voice, and useful prompts. <Google officially launches “Gemini 3”! What features does the “world's smartest AI model” have?> This article was first published in BlockTempo, the most influential blockchain news media.