The latest pre-release version of Google's Gemini 2.5 Pro language model brings major improvements for front-end development and complex programming tasks.
Google has launched an updated preview of its Gemini 2.5 Pro AI model, called the "I/O Edition." The release targets developers interested in testing the model ahead of the upcoming Google I/O conference. The new version is designed to improve performance on everyday programming tasks and support the creation of agent-driven workflows.
Google says Gemini 2.5 Pro is currently the top performer on the WebDev Arena leaderboard—a benchmark that uses human reviewers to judge the look and functionality of web apps, though the ranking itself is somewhat controversial. Startups like Cognition and Replit are already using the updated model to build code agents that can generate apps on their own.
Video analysis and app generation
One new feature is video content analysis. Gemini 2.5 Pro scores 84.8 percent on the VideoMME benchmark and can generate applications directly from video input. The model can also create code-based visualizations from images; here's an example in the Gemini app.
Another demo, "Video to Learning App", turns a single YouTube video into an interactive learning platform, complete with a user interface and instructional structure.
Gemini 2.5 Pro can help automate adding new features to existing web projects. Typically, developers would study design files and write CSS to set up colors, spacing, and fonts. Now, Gemini 2.5 Pro can partially automate this workflow. In one example, a video player styled after the Gemini 95 starter app was added to a project by the model itself.
The dictation app shows how quickly conceptual ideas can be turned into working applications. The interface features animated microphone waves, responsive layouts, and subtle hover effects. Google says Gemini 2.5 Pro blends a solid grasp of design principles with options for fine user control.
Developers can access the new 2.5 model through the Gemini API in Google AI Studio or Vertex AI. Existing users of the previous version (03-25) are automatically routed to the update (05-06). According to Google, the new release reduces function call errors and increases the trigger rate for function actions, with no change in price. An updated model map is also available.
Gemini Pro 2.5 is Google's most powerful AI model so far. As a multimodal reasoning system, it can handle text, audio, images, video, and code. It performs especially well on complex reasoning challenges—such as math and science tests like GPQA and AIME—and excels at programming tasks, particularly in web app development and code transformation.
