Google I/O 2025: Enhanced Gemini 2.5 AI Models with Deep Learning
Google unveiled a series of innovative features for its Gemini 2.5 family of artificial intelligence models during the Google I/O 2025 event on Tuesday. Among the highlights was the introduction of an advanced reasoning mode called Deep Think, powered by the Gemini 2.5 Pro model. Additionally, the tech giant presented Native Audio Output, a new feature that enables more natural and human-like speech through the Live application programming interface (API). These enhancements also include thought summaries and thinking budgets, aimed at improving the experience for developers using the latest Gemini models.
Gemini 2.5 Pro Achieves Top Rankings
In a detailed blog post, Google outlined the new capabilities set to be integrated into the Gemini 2.5 AI model series over the coming months. Earlier this month, the company released an updated version of the Gemini 2.5 Pro, which showcased enhanced coding abilities and secured the top spot on both the WebDev Arena and LMArena leaderboards. The introduction of the Deep Think mode marks a significant improvement, allowing the Gemini 2.5 Pro to evaluate multiple hypotheses before generating a response. This new reasoning mode employs a distinct research technique compared to the Thinking versions of previous models.
Google shared benchmark scores from internal testing, highlighting the impressive performance of the Gemini 2.5 Pro Deep Think. It achieved a score of 49.4 percent on the 2025 UAMO, a challenging mathematics benchmark test, and demonstrated competitive results on LiveCodeBench v6 and MMMU. Currently, Deep Think is undergoing testing, with Google conducting safety evaluations and gathering feedback from safety experts. At this stage, the reasoning mode is accessible only to trusted testers via the Gemini API, and no official release date has been announced.
Enhancements to Gemini 2.5 Flash Model
Google also revealed improvements to the Gemini 2.5 Flash model, which was launched just a month prior. The company stated that key benchmarks for reasoning, multimodality, coding, and long context have all seen enhancements. Furthermore, the updated model is reported to be more efficient, utilizing 20-30 percent fewer tokens than its predecessor. Developers can currently access this new version of Gemini 2.5 Flash in preview mode through Google AI Studio, while enterprises can utilize it via the Vertex AI platform. The model is expected to be widely available for production by June.
In addition to these updates, developers using the Live API will benefit from a new feature within the Gemini 2.5 series. Google introduced a preview version of Native Audio Output, which allows for the generation of speech that is more expressive and human-like. This feature enables users to customize the tone, accent, and style of the generated speech, enhancing the overall interaction experience.
New Features for Developers
The early version of Native Audio Output includes three key functionalities. The first, Affective Dialogue, allows the AI model to detect emotions in a user’s voice and respond appropriately. The second feature, Proactive Audio, enables the model to focus solely on the user, ignoring background conversations until addressed. Lastly, the Thinking feature leverages Gemini’s reasoning capabilities to provide verbal answers to complex queries.
Moreover, the Gemini 2.5 Pro and Flash models will now include thought summaries in the Gemini API and Vertex AI. These summaries reveal the model’s raw thought processes, which were previously exclusive to Gemini’s reasoning models. Google plans to provide detailed summaries that include headers, key details, and information about the model’s actions with each response.
In the upcoming weeks, developers will also gain access to thinking budgets with the Gemini 2.5 Pro. This feature will enable them to manage the number of tokens consumed by the model before it generates a response. Additionally, the Computer Use agentic function from Project Mariner will soon be integrated into the API and Vertex AI, further expanding the capabilities available to developers.
Observer Voice is the one stop site for National, International news, Sports, Editor’s Choice, Art/culture contents, Quotes and much more. We also cover historical contents. Historical contents includes World History, Indian History, and what happened today. The website also covers Entertainment across the India and World.
