Google I/O 2025: Gemini 2.5 AI Models Upgraded With Deep Think Mode
Google showcased several new features for the Gemini 2.5 family of artificial intelligence (AI) models at the Google I/O 2025 event on Tuesday. The Mountain View-based tech giant introduced an enhanced reasoning mode dubbed Deep Think, powered by the Gemini 2.5 Pro model. Another new feature is the Native Audio Output for more natural and human-like speech, which will be accessible through the Live API. Along with these upgrades, thought summaries and thinking budgets are being added to the latest Gemini models for developers.
New Features in Gemini 2.5 Pro Model
In a blog post, Google detailed the new capabilities of the Gemini 2.5 AI models that will be rolled out in the upcoming months. The updated Gemini 2.5 Pro model now includes the Deep Think mode, allowing the AI to consider multiple hypotheses before generating a response. This new mode utilizes a unique research technique compared to the older Thinking versions of the models.
Enhancements in Gemini 2.5 Flash Model
Google has also enhanced the Gemini 2.5 Flash model, released just a month ago, with improved benchmarks for reasoning, multimodality, code, and long context capabilities. The updated model is more efficient, utilizing 20-30 percent fewer tokens while maintaining performance standards. The new version of Gemini 2.5 Flash is currently available in preview for developers via Google AI Studio and will be widely accessible for production in June.
Introduction of Native Audio Output
Developers using the Live API now have access to a preview version of Native Audio Output in the Gemini 2.5 series of AI models. This feature enables the generation of speech in a more expressive and human-like manner, allowing users to control tone, accent, and style. The feature includes Affective Dialogue for emotion detection, Proactive Audio to filter background conversations, and Thinking for complex query responses.
Upgrade in Thought Summaries and Thinking Budgets
In addition to the new features, the Gemini 2.5 Pro and Flash models in the Gemini API and Vertex AI will now display thought summaries, providing insights into the model's raw thought process with details on headers, key information, and model actions. Developers will soon be able to utilize thinking budgets with the Gemini 2.5 Pro to control token consumption before responses are generated. Furthermore, Project Mariner's Computer Use agentic function will also be integrated into the API and Vertex AI in the near future.
For the latest tech news and reviews, follow Gadgets 360 on Twitter, Facebook, WhatsApp, Threads, and Google News. For the latest videos on gadgets and tech, subscribe to our YouTube channel. If you want to know everything about top influencers, follow our in-house Who'sThat360 on Instagram and YouTube.
