/bmi/media/media_files/xSZB6ThNzW4DotyLtq1Y.jpg)
New Delhi: Google has introduced Gemini 2.5 Deep Think, a cutting-edge AI reasoning feature now available to Google AI Ultra subscribers in the Gemini app.
Launched on August 1, 2025, Deep Think employs a multi-agent system to tackle complex queries by exploring multiple solutions in parallel, delivering more accurate and nuanced responses for tasks requiring advanced reasoning, such as coding, mathematical problem-solving, and scientific analysis.
Unlike conventional AI models that provide rapid, single-path answers, Deep Think spawns multiple AI agents to evaluate various approaches simultaneously, refining them to produce optimal solutions. This approach, while resource-intensive, has demonstrated remarkable capabilities, with a variant of the model achieving gold-medal performance at the 2025 International Mathematical Olympiad (IMO).
The consumer version, optimised for daily use, achieves bronze-level IMO performance and excels in benchmarks like LiveCodeBench V6 (87.6%) and Humanity’s Last Exam (34.8%), surpassing competitors like xAI’s Grok 4 (79%) and OpenAI’s o3 (72%) in coding tasks.
Available exclusively to Google AI Ultra subscribers at $250 per month, Deep Think integrates tools like Google Search and code execution, enabling it to handle intricate tasks such as algorithm development, web design, and navigating dense scientific literature. Google CEO Sundar Pichai announced the rollout on X, emphasising its utility for “complex time complexity analysis to deep scientific reasoning,” calling it a tool for enthusiasts tackling challenging problems.
Google is also sharing a high-performance variant with select mathematicians and academics to explore its research applications, with plans to extend access to trusted developers via the Gemini API in the coming weeks.