Stay up to date with the latest news and exclusive content on cutting-edge AI technology by subscribing to our daily and weekly newsletters. Find out more
A revolutionary advancement has emerged in the realm of AI coding models: Google’s DeepMind AI research unit has introduced Gemini 2.5 Pro “I/O” edition, a new iteration of its highly successful Gemini 2.5 Pro multimodal large language model (LLM) released earlier this year. DeepMind CEO Demis Hassabis hailed it as “the best coding model we’ve ever built.”
The latest benchmarks from the company reveal that Google has taken the lead in the generative AI race, surpassing all other models in at least one significant coding benchmark for the first time since the launch of ChatGPT in late 2022.
The new version, named “gemini-2.5-pro-preview-05-06,” has replaced the previous 03-25 release and is now accessible to independent developers on Google AI Studio, enterprises on the Vertex AI cloud platform, and individual users through the Gemini app. According to Google’s blog post, it also powers features in the Gemini mobile app, including Canvas and other functionalities.
This updated version enhances feature development in apps like Gemini 95, facilitating automatic matching of visual styles across components. It also enables tasks like converting YouTube videos into comprehensive learning applications and creating highly styled components, such as responsive video players or animated dictation interfaces, with minimal manual CSS editing.
As a proprietary model, enterprises must pay Google to utilize it and can access it solely through Google’s web services. However, there are no changes to pricing or rate limits; current users of Gemini 2.5 Pro will be automatically transitioned to the updated model, priced at $1.25/$10 per million tokens in/out (for context lengths of 200,000 tokens) compared to Claude 3.7 Sonnet’s $3/$15.
This development, ahead of Google’s upcoming I/O developer conference later this month, is positioned as a response to positive community feedback regarding Gemini’s practical utility in real-world code generation and interface design.
Logan Kilpatrick, Senior Product Manager for Gemini API and Google AI Studio, confirmed in a developer blog post that the update addresses key developer feedback related to function calling, enhancing error reduction and trigger reliability.
Leading Scores in Web App Generation
Gemini 2.5 Pro Preview (05-06) has surpassed Anthropic’s Claude 3.7 Sonnet to claim the top spot on the WebDev Arena Leaderboard, a third-party metric ranking models based on human preference for generating visually appealing and functional web apps.
The new version scored 1499.95 on the leaderboard, outperforming Sonnet 3.7’s 1377.10. The previous Gemini 2.5 Pro (03-25) model held the third position with a score of 1278.96, marking a significant 221-point increase with the I/O edition.
According to AI power user “Lisan al Gaib,” even OpenAI’s GPT-4o (“o3”) failed to displace Sonnet 3.7, underscoring the significance of Gemini’s progress.
The enhanced performance of Gemini reflects advancements in reliability, aesthetics, and usability in its outputs.
Praise for Gemini 2.5 Pro
Several developers and industry leaders have commended the model’s improved reliability and practical application in production settings.
Silas Alberti from Cognition noted that Gemini 2.5 Pro successfully executed a complex refactoring of a backend routing system, showcasing decision-making akin to a seasoned developer.
Michael Truell, CEO of AI coding tool Cursor, reported a significant decline in tool call failures during internal testing, indicating enhanced effectiveness in hands-on environments. Cursor has seamlessly integrated Gemini 2.5 Pro into its code agent, demonstrating how developers are leveraging the model in intelligent workflows.
Michele Catasta, President of Replit, described Gemini 2.5 Pro as a leading model for balancing capability and latency, hinting at potential integration into their tools for tasks requiring high responsiveness and reliability.
AI educator and BlueShell private AI chatbot founder Paul Couvert praised Gemini 2.5 Pro for its impressive code and UI generation capabilities.
Additionally, Pietro Schirano, CEO of AI art tool EverArt, highlighted Gemini 2.5 Pro’s ability to generate interactive simulations swiftly, underscoring its potential in diverse applications.
RameshR (@rezmeram) on X showcased a Tetris-style puzzle game with sound effects created in under a minute, emphasizing the model’s versatility and impact.
These endorsements bolster DeepMind’s claims of practical enhancements and could drive broader adoption across developer platforms.
Creating Full Apps from a Single Prompt
A standout feature of the update is its capability to construct complete, interactive web apps or simulations from a single prompt, aligning with DeepMind’s goal of simplifying prototyping and development processes.
Demos within the Gemini app illustrate how users can translate visual patterns or thematic prompts into functional code, reducing barriers for design-oriented developers and teams exploring new concepts.
While the technical details of Gemini 2.5 Pro’s architecture remain undisclosed, the focus remains on facilitating faster and more intuitive development experiences. Positioned as a practical tool for real-world coding challenges, Gemini 2.5 Pro aims to meet developer demands and maintain momentum ahead of major conference announcements.
