Google recently shipped a major update to its Gemini 2.5 AI model, as DeepMind CEO Demis Hassabis says it’s “the best coding model” that the company has ever built, ranked as the top model in LMArena in coding and the WebDev Arena Leaderboard.
According to Google:
“2.5 Pro now leads on the WebDev Arena Leaderboard, surpassing the previous version by +147 Elo points. This leaderboard measures human preference for a model’s ability to build aesthetically pleasing and functional web apps. It also continues to build on its strong foundation in native multimodality and long context; it has state-of-the-art performance in video understanding, with a score of 84.8% on the VideoMME benchmark.”
Google calls the updated model the I/O edition, referring to the company’s annual I/O developer conference, which is expected to run from May 21 through May 25. There, Google is expected to unveil a host of new AI models.
Very excited to share the best coding model we’ve ever built! Today we’re launching Gemini 2.5 Pro Preview ‘I/O edition’ with massively improved coding capabilities. Ranks no.1 on LMArena in Coding and no.1 on the WebDev Arena Leaderboard.It’s especially good at building… pic.twitter.com/9vRaP6RTToMay 6, 2025
According to the company, the updated AI model ships with improved coding capabilities, making it easier for developers to build interactive web apps. “Beyond UI-focused development, these improvements extend to other coding tasks such as code transformation, code editing and developing complex agentic workflows,” Google added.
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.
Cursor AI CEO, Michael Truell
To that end, Google says the new update will improve Gemini 2.5 Pro’s coding performance and “address key developer feedback including reducing errors in function calling and improving function calling trigger rates.”
The updated model is available via the Gemini API. You can also access it from Google’s Vertex AI and AI Studio. Finally, users can access it directly from the Gemini app, allowing them to “vibe code” and build interactive web apps.