Google DeepMind has announced the release of Gemini 2.5 Deep Think, a cutting-edge AI model that the company claims is its most powerful reasoning system to date.

The model, launching Friday, will be available to users subscribed to Google’s $250-per-month Gemini Ultra plan.

Unveiled at Google I/O in May, Gemini 2.5 Deep Think is the company’s first publicly available multi-agent model—an architecture where multiple AI agents work in parallel to explore and solve problems. While this setup consumes significantly more computing power, it delivers markedly better results, according to Google.

The model has already made a mark in competitive arenas. A variant of Gemini 2.5 Deep Think earned a gold medal at this year’s International Math Olympiad. Google now plans to share that IMO variant with select academics and researchers, noting that the system “takes hours to reason” and could aid complex scientific discoveries.

Google says Deep Think outperforms rival models from OpenAI, xAI, and Anthropic on several benchmarks. It scored 34.8% on the “Humanity’s Last Exam,” a test of interdisciplinary knowledge, compared to xAI’s Grok 4 at 25.4% and OpenAI’s o3 at 20.3%. On LiveCodeBench6, a coding benchmark, Deep Think led with 87.6%, ahead of Grok’s 79% and OpenAI’s 72%.

The model integrates tools like code execution and Google Search, and generates longer, more detailed responses. Google says it developed new reinforcement learning techniques to improve its reasoning and creativity.

Despite strong results, multi-agent systems are expensive to operate, and like xAI’s Grok 4 Heavy, Deep Think remains locked behind premium subscriptions.

Google will soon expand access via its Gemini API, aiming to learn how developers and enterprises might deploy the new system.