Google has released an unexpected upgrade to its Gemini 2.5 Pro AI model, with significant improvements focused on coding capabilities, particularly for web development.
Key Points:
- Google released an improved Gemini 2.5 Pro ahead of its I/O conference, focusing on enhanced coding capabilities
- The updated model now ranks #1 on the WebDev Arena leaderboard for building web applications
- The model is available immediately through the Gemini API, with no price increase
Google is releasing an upgraded version of its Gemini 2.5 Pro model today, weeks ahead of schedule, with significantly enhanced capabilities for developers. The company says overwhelming positive feedback from the developer community prompted them to push out the improvements early rather than waiting for their upcoming I/O conference later this month.
The biggest improvements center around coding capabilities, with the updated model showing particular strength in front-end and UI development. According to Google, the new model now tops the WebDev Arena leaderboard, which measures human preference for AI models that build aesthetically pleasing and functional web applications. This represents a jump of 147 Elo points over the previous version.
🏦 For a small taste of how good our new Gemini 2.5 Pro model is at generating UI elements:
— 👩💻 Paige Bailey (@DynamicWebPaige) May 6, 2025
👇Check out how well it was able to replicate a screenshot of an example bank statement - even creating a similar logo, and preserving all of the text:@GoogleAIStudio pic.twitter.com/VGOMYQzPPx
"We found Gemini 2.5 Pro to be the best frontier model when it comes to 'capability over latency' ratio," said Michele Catasta, President of Replit, in Google's announcement. "I look forward to rolling it out on Replit Agent whenever a latency-sensitive task needs to be accomplished with a high degree of reliability."
The updated model also maintains its multimodal capabilities, scoring 84.8% on the VideoMME benchmark for video understanding. This combination of video comprehension and coding allows for new workflows, such as generating interactive applications directly from video content. Google demonstrated this with a "Video to Learning App" feature that creates an interactive learning application based on a single YouTube video.
Beyond front-end development, Google claims the update addresses key developer feedback by reducing errors in function calling and improving function calling trigger rates. The model is also touted for its ability to transform and edit code, as well as create complex "agentic workflows" — essentially AI systems that can autonomously perform sequences of actions.
Michael Truell, CEO of AI coding tool Cursor, praised the update:
"We're excited about the latest Gemini 2.5 Pro, which builds on its already strong real-world coding capabilities. We're observing internally that the new model has a significant reduction in its failure to call tools, an improvement we believe our users will find makes 2.5 Pro even more effective than before in Cursor."
For developers already using Gemini 2.5 Pro, the update comes with no required changes — Google says the previous iteration (03-25) now points to the most recent version (05-06), making the improved capabilities immediately available at the same price point. The model is accessible through Google AI Studio and Vertex AI for enterprise customers, and it's also powering features in the consumer Gemini app.
This early release appears to be part of Google's strategy to maintain momentum in the increasingly competitive AI space. Earlier today, news broke that OpenAI acquired Windsurf, the popular AI coding app for $3 bilion. Google seems eager to showcase its own advancements ahead of its developer conference. With I/O just weeks away, this release might be just an appetizer for bigger announcements to come.