Google is providing early access to an updated version of its Gemini 2.5 Pro multimodal AI model. Called Gemini 2.5 Pro Preview, the model has “significantly” improved capabilities for coding, especially for interactive web applications. Google said it released the preview model early due to “overwhelming enthusiasm” for the model “so people can start building.” The model was supposed to be unveiled at Google’s I/O developer conference later this month. The updates include enhancing code transformation, code editing and developing of complex agentic workflows. Google said the updates enabled the model to top the chart in the WebDev Arena Leaderboard, which ranks large language models based on their performance in web development. This ranking measures how well a model can build “aesthetically pleasing and functional” web applications. In the arena, AI models compete with each other in front-end UI design and coding contests to earn ELO points. Google said Gemini 2.5 Pro Preview has surpassed the prior version by 147 ELO points. Gemini 2.5 Pro is in third place, after Claude 3.7 Sonnet. OpenAI’s GPT-4.1 came in fourth and Claude 3.5 Sonnet is fifth. Gemini 2.5 Pro Preview also scored well in video understanding, at 84.8% on the VideoMME benchmark, which assesses the capabilities of multimodal models in how well they analyze videos. On WebDev Arena Leaderboard, a third-party metric that ranks models by human preference based on their ability to generate visually appealing and functional web apps, Gemini 2.5 Pro Preview (05-06) has now overtaken Anthropic’s Claude 3.7 Sonnet at the number one spot.