Gemini 2.5 Pro leads WebArena rankings with improved web development performance.
Google has unveiled Gemini 2.5 Pro Preview (I/O edition), its latest AI model update, ahead of the annual I/O developer conference.
The new version introduces upgrades designed to enhance coding assistance, video comprehension, and web development capabilities, further highlighting the company’s commitment to integrating generative AI across its products.
According to Google, Gemini 2.5 Pro delivers noticeable improvements in software development workflows. The AI model now ranks first on the WebArena leaderboard, which assesses the quality of AI-generated websites through human feedback.
Google reports a 147-point increase in Elo score compared to the previous version. Enhanced features also include better code interpretation and generation, real-time function calling, and lower rates of silent failures in tasks such as bug fixing and editing.
Video analysis is another key area of progress. Gemini 2.5 Pro achieved an 84.8% score on the VideoMMe benchmark, which evaluates AI performance in understanding and answering questions about visual content, particularly educational videos.
This marks another step in Google’s multimodal AI ambitions, which seek to unify text, images, code, audio, and video into cohesive AI systems. However, Google has not detailed how privacy and content safety are managed when processing video data.
Developers can now access Gemini 2.5 Pro via the Gemini API, Google AI Studio, and Vertex AI, as well as through Gemini’s mobile and web apps.
Currently released as a preview, the model will likely receive updates following developer feedback during and after the I/O event. Details about general availability or potential pricing changes have not been disclosed.
The release of Gemini 2.5 Pro comes as AI competition intensifies, with OpenAI, Anthropic, and Meta also advancing their foundation models. While Google highlights Gemini’s strong performance in coding and web development, broader comparisons between models remain limited for now.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!