Gemini 2.5 Pro Preview, an updated AI model optimized for coding, is here

What you need to know

• Gemini 2.5 Pro Preview offers improved coding performance and advanced reasoning techniques.
• The model will be officially updated at Google I/O 2025, but is available for early access starting today.
• Users can try out Gemini 2.5 Pro Preview (I/O Edition) in Google AI Studio and Vertex AI starting now.

Google’s experimental versions of Gemini 2.5 Pro are getting a boost, as there’s a new Gemini 2.5 Pro Preview (I/O edition) model available for tasks that involve coding and advanced reasoning. The early access comes ahead of Google I/O 2025, an annual developer conference later this month, the company announced in a blog post. The full update will be released at the event, but the preview is already a top performer on key AI benchmarks.

Users will find “meaningful improvements for front-end and UI development, alongside improvements in fundamental coding tasks such as transforming and editing code, and creating sophisticated agentic workflows,” the company explained in a separate post on the Google for Developers blog.

Gemini 2.5 Pro Preview ranks first on the WebDev Arena leaderboard at the time of publishing, a benchmark that specifically tests AI models’ performance in web development exercises. It tops Claude 3.7 Sonnet, which now holds second place. Google holds two of the top three WebDev Arena leaderboard spots, with Gemini 2.5 Pro Experimental still holding third place.

Aside from Gemini 2.5 Pro Preview’s coding prowess, the AI model is also capable of understanding video. It scores 84.8% on the VideoMME benchmark, and in an example, Google shows off how it can turn a single YouTube video into a complete learning app.

Gemini 2.5 Pro: Video to code capability – YouTube

Watch On