Google has announced significant upgrades to its AI model Gemini 2.5, featuring improved performance, new capabilities, and enhanced developer experience.
Gemini 2.5 Pro is now the world-leading model across WebDev Arena and LMArena leaderboards, and it helps people learn more effectively than ever before. The updated model includes a new “Deep Think” mode that uses advanced reasoning techniques to consider multiple hypotheses before responding.
The 2.5 Flash version of Gemini is also available to everyone in the Gemini app, with a general availability release planned for early June for developers and enterprises in Vertex AI. This version features improved performance, including using 20-30% fewer tokens in evaluations.
New capabilities include native audio output and improvements to the Live API, which allows users to build conversational experiences with more natural and expressive dialogue. The model can also detect emotion in a user’s voice and respond accordingly.
Google has also introduced new security safeguards to protect against security threats like indirect prompt injections. Additionally, the company is extending thinking budgets to Gemini 2.5 Pro for more control over cost by balancing latency and quality.
The developer experience has been improved with the introduction of thought summaries in the Gemini API and Vertex AI, as well as support for Model Context Protocol (MCP) definitions. This allows developers to integrate Gemini models with open-source tools more easily.
Overall, these upgrades demonstrate Google’s continued commitment to advancing its AI technology and providing users with more powerful and intuitive tools.
Source: https://blog.google/technology/google-deepmind/google-gemini-updates-io-2025