Google has unveiled the Gemini 2.5 Pro Preview ahead of its annual developer conference, Google I/O 2025, which is scheduled for later this month. The new AI model, an updated version of the Gemini 2.5 Pro, has been launched earlier than expected. Google aims to offer developers early access to the model’s enhanced features, allowing them to start building with the latest improvements before the official release.

Tulsee Doshi, Senior Director of Product Management at Google, explained that the Gemini 2.5 Pro Preview was initially slated for release at Google I/O, but the company decided to make it available sooner due to high demand from developers. The decision ensures that developers can begin exploring the new features right away.

Gemini 2.5 Pro Preview: Improved Coding and App Building Features

On the other hand, Google DeepMind CEO Demis Hassabis shared the launch of Gemini 2.5 Pro Preview on X, calling it the “best coding model ever built.” He highlighted the model’s potential for creating interactive applications and posted a video demonstrating how it can be used to prototype ideas. Hassabis also emphasised that the new model has improved capabilities, which will make it a top choice for coding tasks.

The Gemini 2.5 Pro Preview offers several enhanced features based on positive feedback about its predecessor. Key improvements include better support for code transformation, code editing, and the development of complex agentic workflows. The updated model also shows progress in user interface development. On the WebDev Arena Leaderboard, the model surpasses its earlier version by 147 Elo points, reflecting its stronger ability to create visually appealing and functional web applications. It also performs well in multimodal tasks, scoring 84.8 percent on the VideoMME benchmark for video understanding.

Gemini 2.5 Pro Preview: How to Access

The Gemini 2.5 Pro Preview is now available for developers through the Gemini API on Google AI Studio and Vertex AI. Additionally, the Gemini app integrates this updated model, which will help users to create and build interactive web applications with a single prompt in Canvas.