On Tuesday, Google unveiled Gemini 2.5 Pro Preview (I/O edition), an enhanced iteration of its premier Gemini 2.5 Pro AI model, which the company asserts surpasses several prevalent benchmarks.

You can now use the Gemini 2.5 Pro Preview (I/O edition) through the Gemini API and on Google's Vertex AI and AI Studio platforms. It costs the same as the Gemini 2.5 Pro model that it replaces. You can also find it in Google's Gemini chatbot app for websites and mobile phones.

The model's launch precedes Google's yearly I/O developer conference (thus the "I/O edition" label), where Google is anticipated to introduce a range of models, alongside AI-driven tools and platforms. The corporation is contending vigorously for dominance in both public perception and market control within the fiercely competitive AI sector; competitors such as OpenAI and xAI are on the brink of launching models projected to exhibit superior performance.

Google asserts that the Gemini 2.5 Pro Preview (I/O edition) exhibits substantially enhanced capabilities for coding and constructing interactive web applications. The company further indicates the model's proficiency in tasks such as code transformation—the modification of code to attain a specific objective—and code editing.

Google highlights in a blog post that Gemini 2.5 Pro Preview (I/O edition) currently holds the top position on the WebDev Arena Leaderboard, a benchmark assessing a model's proficiency in generating visually appealing and effective web applications. Furthermore, Gemini 2.5 Pro Preview (I/O edition) exhibits cutting-edge capabilities in video comprehension, achieving a remarkable 84.8% score on the prominent VideoMME benchmark.

Google stated in its blog post that this new version of Gemini 2.5 Pro will not only enhance coding efficiency for existing users but also tackle significant developer concerns, such as minimising errors in function calling and increasing function calling trigger rates. It added that, by design, the model demonstrates a distinct aptitude for visually appealing web development while remaining controllable.