DeepMind CEO calls Google’s updated Gemini 2.5 Pro AI “the best coding model” with a taste for aesthetic web development

Google recently released a significant upgrade for its Gemini 2.5 AI system. According to DeepMind CEO Demis Hassabis, this latest model is considered the finest programming model that the company has ever constructed. It currently holds the top spot in LMArena for coding and leads the WebDev Arena Leaderboard.

According to Google:

2. Version 2.5 of Pro is currently at the top of the Web Development Arena Leaderboard, outperforming its previous version by a significant margin of 147 Elo points. This leaderboard assesses user preference for a model’s capacity to create visually appealing and functional web applications. It also reinforces its robust base in native multimodality and long-term context, demonstrating cutting-edge capabilities in video comprehension, with an impressive score of 84.8% on the VideoMME benchmark.

Google has labeled the latest version as the I/O Conference Edition, since

As per the company’s announcement, the revised AI model now offers enhanced programming abilities, simplifying the process for developers to construct dynamic web applications. Google further explained that these improvements go beyond user interface development, extending to tasks like code transformation, editing, and creating sophisticated workflows involving intelligent agents.

We’re thrilled about the upgraded Gemini 2.5 Pro, which further strengthens its robust programming abilities in real-world scenarios. Internally, we’ve noticed a substantial decrease in the tool misfire issues with this new version, an improvement we think will make the 2.5 Pro even more efficient than before, particularly in the Cursor environment.

For this purpose, Google has announced that the latest update for Gemini 2.5 Pro will enhance its coding efficiency. Moreover, it aims to rectify common issues related to function calling, such as reducing errors, and boosting the rate at which functions are triggered.

The revised model can be obtained through the Gemini API. It can also be accessed on Google’s Vertex AI and AI Studio platforms. Furthermore, users have the option to access it directly within the Gemini app, enabling them to “code along” and create interactive web applications.

Read More

2025-05-07 16:25