“Gemini热潮”再度席卷全球！谷歌Deep Think“硬核升级”直指大型科研工程

Core Insights - Google has significantly upgraded its Gemini 3 AI model with the Deep Think mode, focusing on addressing complex challenges in modern scientific research and engineering, sparking a new wave of interest in "Gemini AI" [1][6] - The upgraded Deep Think mode is now available to Google AI Ultra subscribers and is the first time Google has provided this functionality through the Gemini API to select researchers, engineers, and large enterprises [1][7] Performance Metrics - The Deep Think mode achieved a score of 48.4% on the Humanity's Last Exam, setting a new industry standard for AI models [4][6] - It scored 84.6% on the ARC-AGI-2 benchmark, verified by the ARC Prize Foundation, and obtained a 3455 Elo rating on the Codeforces competitive programming platform [4][6] - The model also demonstrated gold medal-level performance in the written sections of the 2025 International Physics and Chemistry Olympiads, achieving 50.5% on the CMT-Benchmark [3][5] Application and Functionality - Deep Think is designed to facilitate practical applications, enabling researchers to interpret complex data and engineers to model intricate physical systems through coding [2][6] - The model's capabilities extend beyond mathematics and programming, now encompassing interdisciplinary research problems requiring a combination of physical intuition, structured chemical inference, mathematical formalization, and coding solutions [3][5] - The upgrade emphasizes structured reasoning and scalable inference capabilities, allowing for iterative reasoning across multiple hypothesis spaces [6][7] Market Position - Google’s Gemini 3 Deep Think competes with AI products from OpenAI and Anthropic, positioning itself as a leading solution in the AI landscape [5] - The model's ability to convert sketches into 3D printable files enhances its appeal and potential for widespread adoption in enterprise research and development workflows [7]