Google Gemini Introduces Deep Think Reasoning Model for 2.5 Pro and better 2.5 Flash
Google started testing a Reasoning Model for Deep Think Gemini 2.5 ProThe company revealed its I/O developer meeting. According to the Deepmind CEO Demis Hasabis, Gemini’s Deep Think uses “fresh cutting-edge research”, which gives the model the ability to consider multiple hypotheses before responding to questions. Google said the “impressive score” was awarded when the 2025 United States of America’s Mathematical Olympiad competition was evaluated. However, Google wants to take more time to get more input from security professionals before maintaining security evaluation and widely released. Therefore, it is therefore a deeper idea for trusted tests by Gemini API to obtain their opinion first.
The company also introduced a better Gemini 2.5 Flash modelIt is optimized for speed and capacity. It is now more effective than the previous one, using fewer tokens and score more in the benchmarks for logical, multimodiality, code and long context than its ancestors. It is usually available in early June. At the moment, the Gemini is previewed by Google AI Studio for better Gemini 2.5 flash developers, by Vertex AI for Enterprise Customers and Gemini app for other users.
Most efficiency profits covered in the I/O stage are focused on 2.5 flash, and Google has announced that it will bring the 2.5 flash concept of “Thinking Budgets” to its sophisticated 2.5 Pro model. This feature allows you to balance the tokens vs. accuracy and the speed of the output.
Separately, Google is bringing Project Mariner Gemini also into API and vertex AI. Project mariner Google’s Gemini-powered AI agents, which can navigate pages in the web browser to complete the tasks for users. This summer is more widespread by the company agents, thereby experimenting with developers. In addition, the company is releasing new previews for text-to-speech on both 2.5 Pro and 2.5 Flash Models by Gemini API, with support of two voices in 24 languages.
