Google releases Gemini 2.5 Deep Think for AI Ultra subscribers

Today, Google has unveiled its most advanced AI model yet, Gemini 2.5 Deep Think, but access is limited to select users. Initially showcased at the I/O conference in May, this powerful AI is now available exclusively through the Gemini app for subscribers of Google's premium $250 AI Ultra plan. Gemini 2.5 Deep Think is engineered to tackle the most intricate queries, utilizing significantly more computational resources than its predecessors. This model builds on the foundation laid by Gemini 2.5 Pro but enhances its capabilities by increasing the duration for analysis, allowing for more comprehensive parallel processing. According to Google, Deep Think is designed to explore various strategies in problem-solving, capable of revisiting and reworking multiple hypotheses it generates. This iterative process is aimed at producing superior results. Notably, Deep Think requires several minutes to formulate responses, which contributes to its improved performance in areas such as design, scientific reasoning, and coding. Google has subjected it to rigorous testing against benchmarks, demonstrating that it outperforms both the standard Gemini 2.5 Pro and competing models like OpenAI's o3 and Grok 4. One standout achievement for Deep Think is its performance on Humanity's Last Exam, a challenging set of 2,500 complex, multi-modal questions spanning over 100 subjects. While other models typically score between 20 to 25 percent, Gemini 2.5 Deep Think achieved an impressive score of 34.8 percent, showcasing its capabilities in handling demanding inquiries.