Google DeepMind is introducing Gemini 2.5 Deep Think, which it claims is its most advanced AI reasoning model. This model can answer questions by exploring multiple ideas simultaneously, ultimately selecting the best response.
Starting Friday, subscribers to Google’s $250-per-month Ultra subscription will have access to Gemini 2.5 Deep Think through the Gemini app.
Unveiled in May at Google I/O 2025, Gemini 2.5 Deep Think is the company’s first publicly available multi-agent model. Unlike traditional single-agent systems, these multi-agent systems employ several AI agents to address questions in parallel, requiring significantly more computational resources but generally yielding better results.
A variation of Gemini 2.5 Deep Think recently earned a gold medal at the International Math Olympiad (IMO). Alongside this model, Google is releasing a version used at the IMO to a select group of mathematicians and academics. This advanced AI model takes hours to reason, contrasting with the quicker response times of most consumer-facing AI models. Google aims to enhance research efforts with feedback on this multi-agent system.
The company highlights that Gemini 2.5 Deep Think represents a significant upgrade from earlier versions announced at I/O. Google has also developed new reinforcement learning techniques to improve the model’s reasoning capabilities.
“Deep Think can assist in tackling problems that require creativity, strategic planning, and gradual improvements,” the company stated in a blog post shared with TechCrunch.
Gemini 2.5 Deep Think achieved state-of-the-art performance on Humanity’s Last Exam (HLE), scoring 34.8% without tools. In comparison, xAI’s Grok 4 scored 25.4%, and OpenAI’s o3 scored 20.3%. Additionally, the model outperformed those from OpenAI, xAI, and Anthropic on LiveCodeBench6, a challenging coding test, with Gemini scoring 87.6%.
ICYMI: UCC D-Hub launches D-Hub Learn to boost Entrepreneurial Skills
The model integrates seamlessly with tools like code execution and Google Search, producing longer and more detailed responses than traditional AI models. In tests, it generated more aesthetically pleasing web development tasks compared to its competitors, suggesting it could accelerate research and discovery.
Several leading AI labs are now exploring multi-agent approaches. Elon Musk’s xAI has released its own multi-agent system, Grok 4 Heavy, while OpenAI and Anthropic are also developing multi-agent solutions.
Despite their strong performance, multi-agent systems are more expensive to operate than traditional models, leading companies to place them behind higher subscription tiers, as seen with Google and xAI.
In the coming weeks, Google plans to share Gemini 2.5 Deep Think with a select group of testers via the Gemini API to better understand its applications for developers and enterprises.
SOURCE: TECH CRUNCH
