We're introducing a new mode called Deep Think for Gemini 2.5 Pro, which enhances model performance by allowing more time for reasoning and thinking.

The Gemini models are fundamentally trained to know when to use search, enabling them to act like agents that can reason and utilize tools effectively.

The Gemini Developer API introduced many new features, including thought summaries directly in the API, native audio output for better voice control, and a URL context tool that allows models to analyze web links, enhancing the building blocks for new capabilities.

We're introducing thinking budgets in Gemini 2.5 Pro, giving users control over cost and latency versus quality in model responses.

The Gemini 2.0 Pro model has impressed me with its capabilities, particularly its ability to process large amounts of information efficiently.

With the Gemini 2.5 Pro model in Android Studio, we can now provide quick responses for code completion and improve the quality of developer workflows.

Gemini 2.5 is our most secure model yet, with strengthened protections against security threats and improved transparency through thought summaries.

Gemini 2.5 Pro is our most intelligent model yet, and it excels in coding, leading on all major coding leaderboards.

Currently, Gemini excels in whole codebase indexing and planning, while Sonet 3.7 is the best for implementing code changes, although I found GPT 4.1 less effective recently.