Google Enhances Coding Features in Gemini 2.5 Pro Preview Before I/O Developer Conference

Google LLC has just released the early access version of Gemini 2.5 Pro Preview, an updated version of its flagship AI model featuring significantly enhanced coding capabilities.

According to Google, the original plan was to announce the Gemini 2025 Pro Preview model at the upcoming Google I/O 2.5 event in a few weeks. However, due to overwhelmingly positive feedback, the company decided to release it ahead of schedule.

Google mentioned that the previous version of Gemini 2.5 Pro, released in March, was built with chain reasoning, making it the "smartest" model the company has ever launched. This model is one of the most capable in coding and constructing user interfaces using its multimodal understanding and reasoning functions.

"In addition to UI-focused development, these improvements also extend to other coding tasks such as code conversion, code editing, and developing complex agent workflows," said Tulsee Doshi Senior, Google's Director of Product Management.

With these enhancements, the model now leads the WebDev Arena leaderboard, surpassing the previous version's score by 147 points. The ranking evaluates human preference for the model's ability to build powerful yet easy-to-understand web applications.

Gemini 2.5 Pro Preview maintains an industry-standard large context window of 1 million tokens, allowing it to process vast datasets, including large documents and codebases, approximately 1 hour of video or 11 hours of audio. The company plans to extend this window to 2 million tokens.

"We've observed internally a significant reduction in cases where the new model cannot call tools, and we believe our users will find this improvement makes 2.5 Pro more effective than before in Cursor," said Michael Truell, CEO of Anysphere Inc., the developer behind the Cursor AI-powered coding editor.

Google has made the AI model preview immediately available to developers through Google AI Studio and Vertex AI, enabling them to start using the Gemini Application Programming Interface. Users can also access it via the Gemini app, where they can use Canvas to interactively build applications. This allows developers to collaborate with the chatbot while creating interactive web applications using HTML, CSS, and React JavaScript.