Google launches Gemini 2.5 Pro ahead of I/O with new AI upgrades

Google has introduced Gemini 2.5 Pro Preview (I/O edition), the latest update to its AI model line-up, ahead of its annual I/O developer conference. The new version includes technical improvements in coding assistance, video comprehension, and web development tasks.The launch reflects Google’s AI priorities under CEO Sundar Pichai, who has been vocal about integrating generative AI across products.

Focus on coding and web development

According to a blog post from Google, Gemini 2.5 Pro has shown improved performance in software development workflows. The model now ranks first on WebArena, an evaluation leaderboard that measures the quality of AI-generated websites through human feedback. Google reports that Gemini 2.5 Pro improved its Elo score by 147 points compared to its predecessor, although external comparative evaluations are still limited.

The model also features upgrades for interpreting and generating code, particularly in tasks like code editing, bug fixing, and function generation. Google claims that the latest version performs more consistently in real-time function calls and shows a reduced rate of silent failures in coding tasks.

Native video understanding

Gemini 2.5 Pro is also equipped with capabilities for native video processing. It achieved a score of 84.8% on the VideoMMe benchmark, which tests how well AI systems can analyse and answer questions about visual content. The benchmark is designed to assess AI comprehension of video-based learning materials, among other use cases.

This advancement is part of Google’s broader focus on multimodal AI—systems that can interpret and generate responses across text, images, code, audio, and now video. The company has not specified how the model handles privacy or content safety within video interpretation, which remains an open question for researchers and developers.

Availability and access

Gemini 2.5 Pro is now accessible through several Google platforms, including the Gemini API, Google AI Studio, and Vertex AI. It is also integrated into the Gemini mobile and web apps. The release is considered a preview, suggesting further refinements may follow after developer feedback is collected during and after the I/O event.

Developers can begin using the model in supported environments with Google Cloud infrastructure. No information has been shared on broader public release timelines or commercial pricing changes, if any.

Competitive context

The launch of Gemini 2.5 Pro comes amid increasing competition in the AI development space. OpenAI, Anthropic, and Meta have all released updated foundation models in recent months. While Google emphasises Gemini’s coding and web development strengths, independent comparisons between models remain limited.