Google is set to expand its AI capabilities with the upcoming release of Gemini 2.5 Pro, scheduled to launch in early June following the general availability of Gemini 2.5 Flash.
The company recently announced that while Gemini 2.5 Flash is already accessible to everyone through the Gemini app, both models will be made generally available for production use in Google AI Studio for developers and Vertex AI for enterprises in the coming weeks. This rollout represents a significant advancement in Google's AI offerings.
A standout feature of the upcoming Gemini 2.5 Pro will be Deep Think, an experimental enhanced reasoning mode specifically engineered for highly complex mathematical and coding challenges. According to Google DeepMind, Deep Think employs cutting-edge research techniques that enable the model to consider multiple hypotheses before responding. This approach has already demonstrated impressive results, with the model achieving top scores on difficult benchmarks including the 2025 USAMO math competition and LiveCodeBench for competitive coding.
Security has been a major focus for the Gemini 2.5 family. Google has implemented advanced safeguards that significantly increase protection against indirect prompt injection attacks during tool use. These attacks occur when malicious instructions are embedded into data that an AI model retrieves. Through a combination of model hardening, input/output checks, and system-level guardrails, Google has made Gemini 2.5 their most secure model family to date.
Beyond security and reasoning capabilities, both 2.5 Pro and 2.5 Flash will include new features such as native audio output for more natural conversations, thought summaries for improved transparency, and Project Mariner's computer use capabilities. The 2.5 Pro version will also support configurable thinking budgets, allowing developers to control the computational resources used by the model.
Google is taking a cautious approach with Deep Think, conducting additional frontier safety evaluations and gathering input from safety experts before making it widely available. Initially, the feature will be accessible to trusted testers via the Gemini API to collect feedback.
This release continues Google's rapid pace of AI innovation, with Gemini 2.5 Pro already leading on multiple benchmarks for reasoning, multimodality, code generation, and long-context understanding.