Google has pushed early access to a souped-up Gemini 2.5 Pro Preview ahead of their I/O 2025 developer conference in a couple of weeks.
Why the hurry? Well, Google put it down to “overwhelming enthusiasm” and the “amazing things” developers were already cooking up with the previous version of Gemini 2.5 Pro. It seems they’re champing at the bit to get these beefed-up capabilities into the wild.
This isn’t just a fresh lick of paint. The company is championing this new version based on the “overwhelmingly positive feedback” received for the original Gemini 2.5 Pro, especially when it came to coding and its ability to juggle different types of information.
We’re told to expect “meaningful improvements for front-end and UI development,” which will be music to the ears of many web developers. But the good news doesn’t stop there; the enhancements reportedly reach deep into the coding toolkit, covering tricky code transformations, detailed code editing, and even the development of those clever agentic workflows – AI systems that can pretty much run tasks on their own.
The proof, as they say, is in the pudding, or in this case, the leaderboards. Gemini 2.5 Pro has now snatched the top spot on the WebDev Arena leaderboard, outperforming its predecessor by a hefty +147 Elo points. For those not keeping score, this leaderboard is a big deal; it’s where human experts rate how well AI models can build websites that not only work smoothly but also look the business.
This leap on the leaderboard suggests a genuine step forward in AI’s ability to help create truly polished, user-friendly web experiences. Google also mentions that this leading edge is already “powering Cursor’s innovative code agent” and fuelling collaborations with bright sparks at companies like Cognition and Replit, all pushing together at “the frontiers of agentic programming.”
Michael Truell, CEO of Cursor, said: “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.”
Michele Catasta, President of Replit, added: “We found Gemini 2.5 Pro to be the best frontier model when it comes to ‘capability over latency’ ratio. 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.”
According to Silas Alberti from the founding team of Cognition, the updated Gemini 2.5 Pro “achieves leading performance” on its junior-dev evals. It was the first-ever model that solved one of its evals “involving a larger refactor of a request routing backend.”
Alberti says the updated Gemini 2.5 Pro feels “like a more senior developer because it was able to make correct judgement calls and choose good abstractions.” This suggests the AI isn’t just churning out code; it’s starting to show a more nuanced understanding of development challenges.
Beyond the hardcore coding, Gemini 2.5 Pro still flexes its muscles with its ability to understand different types of media and handle a ton of information in one go. Google is particularly proud of its “state-of-the-art performance in video understanding,” and they’ve got the numbers to back it up: an 84.8% score on the VideoMME benchmark. Marry this video savvy with its coding skills, and you get some genuinely new possibilities.