We've seen developers do amazing things with Gemini 2.5 Pro, so we decided to release the updated version a few weeks early to get it into developers' hands faster.
Today we are excited to release Gemini 2.5 Pro Preview (I/O Edition). This update offers even greater coding capabilities that you can start using before Google I/O later this month. Expect significant improvements in front-end and UI development, as well as improvements in core coding tasks such as transforming and editing code and creating sophisticated agent workflows.
“We found the Gemini 2.5 Pro to be the best pioneer model in terms of performance-to-latency ratio. I look forward to implementing it in Replit Agent whenever a latency-sensitive task needs to be performed with a high degree of reliability.”
– Michele Catasta, chairman, Repeat
Best-in-class frontend website development
Gemini 2.5 Pro currently holds the top spot on the list WebDev Arena Leaderboardwhich measures people's preferences for a model's ability to create aesthetically pleasing and functional web applications. Building on these leading capabilities, Gemini 2.5 Pro supports the innovative Cursor code agent and strengthens our partnerships with companies such as Cognition and Replit. Together, we push the boundaries of agent-based programming to unlock new possibilities for developers.
“The updated Gemini 2.5 Pro achieved leading performance in our junior developer tests. It was the first model to solve one of our evaluations that included a larger request routing backend refactorer. I felt like a senior developer because he was able to make good judgment calls and choose good abstractions.”
– Silas Alberti, Founding Team, Cognition
Gemini 2.5 Pro in action
Gemini 2.5 Pro's in-depth code understanding combined with advanced reasoning makes Gemini 2.5 Pro continue to be the model of choice for developers. We are particularly excited to use this model in the following cases.
Video for code
Gemini 2.5 Pro delivers state-of-the-art video understanding, achieving a score of 84.8% in the VideoMME test. Combining this with coding enables new flows that were not previously possible in previous versions. For example Video for an educational application in Google AI Studio shows how Gemini 2.5 Pro creates an interactive educational application based on a single YouTube video. With improved video understanding and a complete user interface, the updated Gemini 2.5 Pro model delivers a more functional experience than the previous simple example.
Easier feature development
Gemini 2.5 Pro excels at front-end website development, so you can get more done. Implementing new features means manually digging into design files and checking components for matching style properties such as colors, fonts, padding, margins, and borders, and then manually writing the CSS code needed to accurately reproduce these visual properties. Now imagine that you are running Gemini 2.5 Pro in an IDE and that the model generates new features such as adding a video player in the style of other applications in the environment Gemini 95 starter application.
Quick concepts of working applications
Bringing your ideas to life with both functionality and a beautiful user interface is easier with Gemini 2.5 Pro. New dictation starter appbuilt using an updated model, it is a perfect example of this in action. Notice some details like wavelength animations, responsive design, and subtle hover effects. By default, the model has a real taste for aesthetically pleasing web development while remaining manageable, helping developers quickly translate a concept into a working web application. Gemini 2.5 Pro was able to design and code a microphone interface animation for the dictation startup application.
Start building with Gemini 2.5 Pro
You can build using Gemini 2.5 Pro with the Gemini API Google Artificial Intelligence Studioand corporate clients can use them Apex AI.
For developers already using Gemini 2.5 Pro, this new version will not only improve coding performance, but also incorporate key developer feedback, including reducing function call errors and improving the trigger rate of function calls. The previous iteration (03-25) now points to the latest version (05-06), so no action is required to use the improved model and it is still available in the same price. We also have I updated the model card with the new version 2.5 Pro.
We can't wait to see the amazing apps you create!
