TLDR: Google has launched Gemini 2.5 Deep Think, an advanced multi-agent AI model designed to tackle complex reasoning tasks by processing multiple ideas simultaneously. This new model, available to Google Ultra plan subscribers, has demonstrated superior performance in benchmarks like the International Math Olympiad and Humanity’s Last Exam.
Google has officially rolled out Gemini 2.5 Deep Think, its most sophisticated artificial intelligence model to date, marking a significant advancement in complex reasoning capabilities. Unveiled at Google I/O in May 2025, this multi-agent AI system is engineered to process and synthesize multiple ideas concurrently, leading to more accurate and comprehensive solutions.
The core innovation of Gemini 2.5 Deep Think lies in its ability to deploy multiple AI agents in parallel on a single problem. This ‘parallel thinking’ approach allows the model to explore various possibilities, brainstorm, and refine ideas over an extended ‘thinking time,’ which can span hours for highly complex calculations, a departure from the seconds or minutes typical of conventional AI models. Google states that this method, while more computationally intensive, yields significantly superior results.
Demonstrating its prowess, a variant of Gemini 2.5 Deep Think recently secured a gold medal at the International Math Olympiad (IMO) 2025, solving intricate mathematical problems. The publicly available version, designed for day-to-day use, still achieves a Bronze-level performance on the IMO 2025 benchmark, according to internal evaluations. Furthermore, in the challenging Humanity’s Last Exam (HLE), Gemini 2.5 Deep Think scored an impressive 34.8% without tools, outperforming competitors such as xAI’s Grok 4 (25.4%) and OpenAI’s o3 (20.3%). In programming challenges, specifically LiveCodeBench6, Gemini 2.5 Deep Think achieved 87.6%, surpassing Grok 4’s 79% and o3’s 72%.
Google is making Gemini 2.5 Deep Think accessible to subscribers of its Ultra plan, priced at $250 per month, within the Gemini app. Users can activate the ‘Deep Think‘ toggle when using Gemini 2.5 Pro. The company is also sharing the full IMO-winning model with a select group of academics and researchers to explore its potential in high-level research tasks. Google plans to extend access to select testers via the Gemini API to understand its applications in business and development, further solidifying its leadership in AI.
Also Read:
- Google AI Unveils Test-Time Diffusion Deep Researcher (TTD-DR) for Advanced AI-Powered Research
- University of Minnesota Adopts Google’s Gemini Generative AI for Systemwide Use
This new model integrates with existing tools like code execution and Google Search, enabling it to provide more detailed and creative responses. Google anticipates that Gemini 2.5 Deep Think will be a transformative tool for fields requiring iterative development, design, and the formulation of complex scientific and mathematical conjectures, potentially accelerating discovery and innovation.
