Google releases Olympiad medal-winning Gemini 2.5 ‘Deep Think’ AI publicly — but there’s a catch…

Google has officially introduced its latest AI model, Gemini 2.5 Deep Think, which is designed for advanced reasoning and intricate problem-solving tasks. This model made headlines recently by winning a gold medal at the International Mathematical Olympiad (IMO), marking a historic achievement as the first AI to do so. However, the version released to the public is not the gold medal-winning model but rather a less powerful 'bronze' variant, according to Logan Kilpatrick, the Product Lead for Google AI Studio. Kilpatrick explained via the social network X, "This is a variation of our IMO gold model that is faster and more optimized for everyday use." While the full gold model is being tested by select mathematicians to assess its comprehensive capabilities, the Gemini 2.5 Deep Think model is now available through the Gemini mobile app, exclusively for subscribers of Google’s premium AI Ultra plan, priced at $249.99 per month. Early adopters have begun exploring the model's capabilities, with mathematicians using it for tasks such as conjecture testing and generating proofs for unsolved problems. For instance, AI expert Ethan Mollick shared his experience with the model, noting its ability to transform a complex prompt into an impressive 3D graphic—a first for any AI model he tested. Google emphasizes the applications of Deep Think, highlighting its superior performance in various benchmark evaluations, including coding and reasoning tests. The model has outperformed competitors like OpenAI’s GPT-4 in several categories, showcasing its enhanced analytical skills and reasoning capabilities. These improvements stem from techniques such as parallel thinking and reinforcement learning, enabling it to tackle nuanced prompts and explore multiple hypotheses more effectively. However, users should be aware that Deep Think's increased sophistication comes with certain trade-offs. Its higher refusal rate may limit its responsiveness to ambiguous queries compared to the standard Gemini 2.5 Pro model. This differentiation allows users to choose between the more agile 2.5 Pro for speedy tasks and the rigorously analytical Deep Think for complex challenges. While the Gemini 2.5 Deep Think model represents a significant leap in AI capabilities, it is essential to note that the released version is a scaled-back iteration of the competition model. Currently, it is available solely through the Google Gemini mobile app for iOS and Android for AI Ultra plan users, who can enable Deep Think by adjusting their settings. In summary, Gemini 2.5 Deep Think serves as a bridge between groundbreaking research and practical application, allowing users to harness the power of a Math Olympiad medal-winning AI. For researchers and enterprises, this model offers a glimpse into the future of collaborative AI in mathematics, while for Ultra subscribers, it marks a significant step toward more advanced and context-aware AI assistance.