Gemini 2.5 gets a massive coding upgrade in the run-up to I/O 2025

Google has announced the early access release of the Gemini 2.5 Pro Preview (I/O edition), an enhanced version of its 2.5 Pro model. This update brings substantial improvements in coding capabilities, particularly in the realm of building engaging and interactive web applications. Originally slated for release at Google I/O in the coming weeks, Google has instead decided to make this preview available sooner in response to the enthusiasm surrounding the latest model.

This latest iteration of Gemini builds upon the already well-received coding and multimodal reasoning features of Gemini 2.5 Pro. Beyond user interface-centric development, these advancements extend to a wider range of coding tasks, including code transformation, code editing, and the development of intricate agentic workflows.

The impact of these new upgrades is evident in Gemini 2.5 Pro’s performance on the WebDev Arena Leaderboard, where it now holds the top position. Surpassing its previous version by an impressive +147 Elo points, this leaderboard serves as a benchmark for human preference regarding an AI model’s ability to construct visually appealing and functional web applications. Furthermore, Gemini 2.5 Pro continues to excel in native multimodality and long context understanding, achieving a state-of-the-art score of 84.8% on the VideoMME benchmark for video understanding.

We’re excited about the latest Gemini 2.5 Pro, which builds on its already strong real-world coding capabilities. We’re observing internally that the new model has a significant reduction in its failure to call tools, an improvement we believe our users will find makes 2.5 Pro even more effective than before in Cursor.

Michael Truell, CEO of Cursor

Developers eager to explore the capabilities of this updated model can begin building through the Gemini API via Google AI Studio and Vertex AI. Additionally, the enhanced Gemini 2.5 Pro is now accessible to users within the Gemini app, powering features such as Canvas. This integration empowers individuals to construct interactive web applications using a single prompt.