At the Google I/O 2025 event on Tuesday, the search giant introduced Gemini 2.5 Pro Deep Think along with a significantly upgraded version of the Gemini 2.5 Flash AI model. The Gemini 2.5 Pro Deep Think mode features advanced reasoning capabilities, “using new research methods that allow the model to evaluate multiple hypotheses before providing a response“.
In the USAMO 2025 test, the Gemini 2.5 Pro Deep Think mode scores 49.4%, outperforming the standard Gemini 2.5 Pro model’s 34.5%. It also surpasses both its predecessor and OpenAI’s o3 model in benchmarks like LiveCodeBench and MMMU. Currently undergoing testing with trusted partners, this advanced mode will be made widely available to users in the future.
Additionally, the new Gemini 2.5 Flash model is significantly smarter. Although smaller and more affordable than the flagship Gemini 2.5 Pro, it delivers impressive performance. On the LMArena leaderboard, Gemini 2.5 Flash ranks just below Gemini 2.5 Pro, with an ELO score of 1424 compared to the Pro model’s 1446.
The new Gemini 2.5 Flash model will be generally available in early June, but you can try the preview version now in the Gemini app, Google AI Studio, and Vertex AI. For developers, it offers enhanced capabilities, greater transparency with thought summaries, and improved cost-efficiency. Plus, you can set a thinking budget for the model.
Additionally, it supports native audio output with the ability to switch between different voices. This audio feature is going live on the Gemini API starting today. Google also states that the Gemini 2.5 Flash model is 22% more efficient and significantly reduces token consumption.
