The article discusses Google’s new “bronze” model available through the Gemini mobile app, a variant of its gold medal-winning model. This version is optimized for daily use and accessible to Google AI Ultra subscribers for $249.99 monthly, with a promotional rate of $124.99/month for the first three months. Google plans to release Deep Think to trusted testers via the Gemini API soon.
The Gemini 2.5 Deep Think model enhances Google’s large language models, focusing on complex problem-solving with capabilities like “parallel thinking” for simultaneous idea exploration and reinforcement learning. It’s intended for tasks requiring extended deliberation, such as mathematical conjecture, scientific research, and algorithm design. Users, including mathematicians, have leveraged it for unsolved problems, and AI experts like Ethan Mollick have praised its unique outputs, such as creating a 3D graphic based on a prompt involving a starship control panel.
Google highlights Deep Think’s application in mathematics, coding, and creative development, obtaining leading scores in benchmark evaluations. The model excels in complex proofs, time complexity, multi-step logic, and creative tasks, outperforming other models like OpenAI’s GPT-4.
Compared to Gemini 2.5 Pro, Deep Think offers superior analytical skills, particularly in nuanced problem-solving, but is slower due to its extensive “thinking time.” Google’s internal comparisons show more refined outputs in creative scenarios.
Despite not being the exact version that won gold at the International Mathematical Olympiad (IMO), Deep Think shares some capabilities with that model. Available through Google’s Gemini app to Ultra subscribers, it offers enhanced AI assistance for enterprise decision-makers, representing a milestone in mathematical AI collaboration.