On Tuesday, Google introduced an upgraded version of its Gemini 2.5 Pro artificial intelligence model, enhancing its coding capabilities. The tech company has named this new iteration the Gemini 2.5 Pro Preview (I/O edition). Initially intended for launch at the Google I/O 2025 event set for May 20-21, the decision to release it ahead of schedule was influenced by the overwhelming positive reception of the model.
Gemini 2.5 Pro Enhanced with Coding Features
In a blog post, Google elaborated on the advancements made to the large language model. The tech giant asserts that the enhanced Gemini 2.5 Pro (I/O edition) boasts a superior understanding of code, enabling it to construct interactive web applications from the ground up. Alongside user interface development, the model is capable of undertaking various intricate coding tasks, including code transformation, editing, and the creation of complex workflows.
The updated model has now claimed the top spot on the WebDev Arena leaderboard, a performance metric devised by LMArena to evaluate user preferences for models completing web development projects. Previously occupying second place, the Gemini model has surpassed Claude 3.7 Sonnet to secure its position as the leader.
An additional feature of the Gemini 2.5 Pro (I/O edition) is its new video-to-code functionality. The model’s ability to interpret videos has contributed to its score of 84.8 percent on the VideoMME benchmark. With its improved coding capabilities, the model can generate interactive web applications based on single YouTube videos; however, this specific feature is presently exclusive to Google AI Studio.
Moreover, enhancements have been made in front-end web development. Google emphasizes that the AI model can now interface with integrated development environments (IDEs) to assess components and ensure consistency in style properties including colors, fonts, padding, and margins.
Developers interested in leveraging the revamped Gemini 2.5 Pro can access it via the Gemini application programming interface (API) through Google AI Studio and Vertex AI. Standard users will find the new model incorporated into the Gemini app and web client, which supports various features, including Canvas.
