Google Upgrades Gemini with 'Deep Think'
Google's Gemini platform has received a major upgrade dubbed "Deep Think." The enhancement reportedly brings more advanced reasoning and contextual understanding to Gemini-powered features across Google Search, Assistant, and Workspace, aiming to handle more nuanced queries.
The "Deep Think" upgrade is a specialized reasoning mode for Google's Gemini 3 AI model, designed to solve complex problems that require more in-depth analysis. It functions by allocating more computation time at the moment of a query, allowing the model to process for longer before delivering a response. This extended reasoning is aimed at tackling challenges in science, research, and engineering. This new mode is being positioned to compete with offerings from other AI labs, such as OpenAI's o1 reasoning models and Anthropic's Claude, which are also focused on advanced analytical capabilities. The development of "Deep Think" is part of a broader industry trend moving beyond general-purpose AI to create specialized engines for professional-grade problems. "Deep Think" has demonstrated significant performance on several rigorous benchmarks. It scored 41.0% on "Humanity's Last Exam" without the use of external tools and 93.8% on GPQA Diamond. On the ARC-AGI-2 benchmark, which measures a system's ability to solve novel challenges, it achieved a score of 45.1%. Access to the "Deep Think" functionality is not available to all users. It is being rolled out as a feature for subscribers of Google AI Ultra, the top-tier plan for Gemini. Google is also making it available to a select group of researchers and enterprises through the Gemini API. This upgrade is part of an ongoing evolution of the Gemini models. The initial Gemini 1 focused on multimodality and a long context window, while Gemini 2 laid the groundwork for agent-like capabilities and advanced reasoning. The Gemini 3 series, which includes "Deep Think," integrates these capabilities to handle more complex, multi-step tasks. The underlying technology of "Deep Think" involves iterative rounds of reasoning and the ability to explore multiple hypotheses simultaneously. This allows it to better handle tasks that require creativity, strategic planning, and step-by-step improvements, such as in algorithmic development and iterative design. For example, it can analyze a sketch and generate a 3D-printable file.
