Google has revealed significant advancements to its top performing Gemini artificial intelligence models, introducing a new mode named Deep Think during its I/O 2025 conference. Deep Think is designed to improve the Gemini 2.5 Pro model by letting it weigh multiple possibilities before answering, boosting accuracy and performance on complex tasks.
Demis Hassabis, who leads Google DeepMind, described Deep Think as using Google’s latest research into advanced reasoning methods and parallel processes. Although specifics remain unclear, comparisons have already been drawn with competing models like OpenAI’s o1-pro due to their similar solution generating approaches.
Upgrades to Gemini 2.5 and More
The improved Gemini 2.5 Pro model set fresh records, outperforming competitors on challenging evaluation platforms such as LiveCodeBench and MMMU. For now, Deep Think is being tested by a small group of trusted users through the Gemini API, with Google holding broader access until more thorough safety checks are completed.
Attention is also being given to the budget friendly Gemini 2.5 Flash, which now brings better ability for coding, handling multiple types of data, logical tasks, and much longer input. This latest version of Flash is more efficient than its predecessor and available for preview in the AI Studio, Vertex AI, and across Google’s Gemini applications.
Developers will see full access to the enhanced Gemini 2.5 Flash in June, according to Google’s timeline. The company has also introduced Gemini Diffusion, a new variant praised for operating four to five times faster than comparable artificial intelligence models of similar size.
Gemini Diffusion is now open for initial testing by select users. These updates reflect Google’s ongoing strategy to stay competitive in the rapidly evolving field of advanced artificial intelligence.
