Alphabet(US:GOOG) Changjiang Securities·2025-11-20 08:48Investment Rating - The investment rating for the industry is "Positive" and is maintained [7] Core Insights - On November 19, Google released Gemini-3 Pro Preview, featuring a sparse MoE architecture that supports up to approximately 1 million tokens for input and a maximum output of 64K tokens. The pricing for input/output is $2/$12 for up to 200K tokens and $4/$18 for more than 200K tokens. The model continues to evolve technically, with an upcoming period of intensive product releases from model vendors overseas. The report emphasizes the importance of monitoring AI product releases that exceed expectations [2][5]. - Gemini 3 Pro is described as the most advanced inference model, significantly outperforming its predecessor Gemini 2.5 Pro and other models like Claude Sonnet 4.5 and GPT-5.1 across major AI benchmarks. It achieved a groundbreaking score of 1501 Elo in LMArena and excelled in various reasoning tasks, showcasing its advanced reasoning capabilities [9]. - The report highlights the transition of AI from "language understanding + knowledge compression" to "autonomous reasoning + multimodal execution," indicating a new phase in AI development. The focus is shifting from model comparison to application comparison, with an emphasis on the potential for AI products to exceed expectations and the progress of AI agents [9]. Summary by Sections Event Description - Google launched Gemini-3 Pro Preview on November 19, featuring a sparse MoE architecture with input support for up to 1 million tokens and a maximum output of 64K tokens. The integration includes Android Studio, Firebase, and Gemini CLI, with specific pricing structures for token usage [5]. Event Commentary - The report discusses the implications of Gemini 3 Pro's capabilities, including its superior reasoning skills and multimodal reasoning performance. It notes the model's ability to understand aesthetic preferences and its potential to transform human-computer interaction. The development experience is described as "agent-first," indicating a shift towards collaborative AI development [9].