Google has significantly expanded its AI capabilities with major updates to the Gemini 2.5 model family announced at its annual I/O developer conference on May 20, 2025.
The most notable advancement is Deep Think, an experimental enhanced reasoning mode for Gemini 2.5 Pro that allows the model to consider multiple hypotheses in parallel before responding. This approach has yielded impressive results, with Google DeepMind CEO Demis Hassabis describing it as "pushing model performance to its limits" using "cutting-edge research in thinking and reasoning."
Deep Think has already demonstrated exceptional performance on challenging benchmarks, scoring impressively on the 2025 USAMO math competition, leading on LiveCodeBench for competitive programming, and achieving 84% on MMMU for multimodal reasoning tasks. Due to its frontier capabilities, Google is initially limiting access to trusted testers while conducting additional safety evaluations.
Google also announced that Gemini 2.5 Flash is now available to everyone in the Gemini app, bringing improved performance on reasoning, coding, and multimodal tasks while using 20-30% fewer tokens. The updated version will be generally available to developers in Google AI Studio and enterprises in Vertex AI by early June, with 2.5 Pro following shortly after.
To provide greater transparency and control, Google is introducing thought summaries in the Gemini API and Vertex AI, organizing the model's reasoning process into a structured format. The company is also extending thinking budgets to 2.5 Pro, allowing developers to control how deeply the model thinks before responding.
These updates coincide with Google's restructuring of its AI subscription tiers. The former AI Premium is now rebranded as Google AI Pro ($19.99/month), while a new Google AI Ultra tier ($249.99/month) offers the highest usage limits, early access to experimental features including Deep Think mode when it launches, and exclusive access to advanced tools like Veo 3 for video generation with native audio.
For developers, Gemini 2.5 Pro remains Google's most advanced model for complex tasks, now leading on both the WebDev Arena and LMArena leaderboards. The company has also integrated Gemini 2.5 into Google Search, with AI Mode now rolling out to all US users and both AI Mode and AI Overviews soon to be powered by a custom version of Gemini 2.5.