The recently upgraded GPT-4 Omni model from OpenAI has been making waves in the AI community. According to Sam Altman, the CEO of OpenAI, this new version of GPT-4 Omni boasts significant improvements in several key areas:

1. Complex Instructions: The model can now handle multi-part prompts with greater precision and consistency, making it more adept at understanding and executing complex tasks.

2. Coding Capabilities: The model has been enhanced for coding, with improved debugging, architectural planning, and the ability to solve challenging coding problems, making it a sharper Dev co-pilot.

3. Intuitive and Creative Thinking: The model generates more original and insightful ideas, making it a valuable tool for brainstorming and reasoning-heavy tasks.

4. Reduced Emoji Usage: The model now focuses more on text-based generation, with a toned-down use of emojis.

5. Improved Coding Instruction Following and Freedom: The new GPT-4 Omni is particularly adept at following coding instructions and offers more freedom in its generation, a significant upgrade from OpenAI's previous restrictions.

The benchmark results from LM Arena have been particularly impressive, with the GPT-4 Omni model surpassing GPT-4.5 and even tying for the top spot on the coding hard prompt benchmark. While it is slightly behind the Gemini 2.5 Pro model, the GPT-4 Omni offers a more cost-effective solution, making it an appealing option for many users.

In this section, we will conduct a side-by-side comparison of the performance of the GPT-4 Omni and Gemini 2.5 Pro models on various prompts and tasks.

Responsive Web App for Income/Expense Tracking

Both models were tasked with building a responsive web application using HTML, CSS, and JavaScript that allows users to track their monthly income and expenses. The application should have features like adding, editing, and deleting transactions, as well as a visual representation of the data.

The GPT-4 Omni model was able to generate a functional application with a sleek design, including a dark mode and the ability to add, edit, and visualize transactions. In contrast, the Gemini 2.5 Pro model produced a responsive front-end but lacked the functionality and features of the GPT-4 Omni's generation.

Result: The GPT-4 Omni model outperformed the Gemini 2.5 Pro in this task, generating a more complete and functional application.

TV Channel Changer

The models were asked to code a TV application that allows users to change channels using the number keys 0-9, with a unique idea for a channel for each number.

The Gemini 2.5 Pro model generated a more comprehensive solution, including a static TV frame and unique animations for each channel. The GPT-4 Omni model was able to generate the channel ideas but did not include the TV frame or animations.

Result: The Gemini 2.5 Pro model performed better in this task, providing a more complete and visually appealing solution.

SVG Butterfly

Both models were tasked with creating an SVG representation of a butterfly with symmetrical wings and simple styling.

Both models performed well in this task, generating clean and visually appealing SVG butterflies. However, the Gemini 2.5 Pro's generation was slightly more polished and visually striking.

Result: Both models passed this task, but the Gemini 2.5 Pro's generation was slightly more impressive.

Tetris Game

The final task was to create a Tetris game in a single HTML file.

Both models were able to generate functional Tetris games, with the GPT-4 Omni's version having a slightly more appealing visual design.

Result: Both models passed this task, with the GPT-4 Omni's generation being slightly more visually appealing.

Overall, the comparison shows that both the GPT-4 Omni and Gemini 2.5 Pro are highly capable models, with each excelling in different areas. The GPT-4 Omni demonstrated stronger performance in generating functional applications with complex features, while the Gemini 2.5 Pro showed superior capabilities in visual and creative tasks, such as the TV channel changer and SVG butterfly.

Both the GPT-4 Omni and Gemini 2.5 Pro models were tasked with building a responsive web app that allows users to track monthly income and expenses. The app should have features like adding, editing, and deleting transactions, as well as a visual representation of the data.

The GPT-4 Omni model generated a sleek and functional app, complete with a dark mode and the ability to add, view, and visualize income and expenses. The app was fully functional and demonstrated the model's capabilities in handling detailed instructions and delivering a practical solution.

In contrast, the Gemini 2.5 Pro model was able to create a responsive front-end, but lacked the functionality of the GPT-4 Omni app. The Gemini model's app did not have a working dark mode or the ability to visualize the income and expense data.

Overall, the GPT-4 Omni model outperformed the Gemini 2.5 Pro in this benchmark, showcasing its superior capabilities in generating a fully functional and feature-rich web application. The GPT-4 Omni's ability to follow complex instructions and deliver a practical solution with attention to detail makes it the clear winner in this head-to-head comparison.

Both the GPT-4 Omni and Gemini 2.5 Pro models were able to generate code for a TV application with channel switching functionality. However, the Gemini 2.5 Pro model demonstrated a clear advantage in terms of visual appeal and interactivity.

The Gemini 2.5 Pro model generated a more visually appealing TV frame, with a well-structured layout and a consistent design across the different channels. The model was also able to create unique animations and visuals for each channel, making the experience more engaging and dynamic.

In contrast, the GPT-4 Omni model was able to generate the basic functionality of channel switching, but the visual representation was more limited. The model did not create a cohesive TV frame or unique visuals for each channel.

When it came to interactivity, the Gemini 2.5 Pro model outperformed the GPT-4 Omni. The Gemini model's TV application allowed users to seamlessly switch between channels, with each channel displaying its own unique content. The GPT-4 Omni model, while functional, lacked the same level of interactivity and responsiveness.

Overall, the Gemini 2.5 Pro model demonstrated a stronger capability in generating a visually appealing and interactive TV application with channel switching functionality, showcasing its superior performance in this particular task.

Both the GPT-4 Omni and Gemini 2.5 Pro models demonstrated impressive capabilities in generating a symmetrical SVG representation of a butterfly. The task of creating an SVG with clean styling and symmetrical wings is a challenging one for large language models, but both models were able to rise to the occasion.

The GPT-4 Omni model produced a visually appealing SVG, with well-balanced wings and a simple, elegant design. The Gemini 2.5 Pro, on the other hand, generated an SVG that the author personally preferred over the GPT-4 Omni's output. Both models were able to successfully capture the essence of a butterfly in their SVG representations, showcasing their artistic prowess and attention to detail.

Ultimately, both models received a passing grade for this prompt, as they were able to meet the requirements of creating a symmetrical butterfly SVG with clean styling. The slight edge goes to the Gemini 2.5 Pro, but the GPT-4 Omni also delivered a strong performance, demonstrating its ability to handle complex artistic tasks.

Both the GPT-4 Omni and Gemini 2.5 Pro models demonstrated impressive capabilities in generating a functional Tetris game within a single HTML file. The generated code from both models was able to create a playable Tetris game, showcasing their proficiency in handling complex coding tasks.

The GPT-4 Omni model produced a visually appealing Tetris game, with a clean and intuitive interface. The game mechanics were well-implemented, allowing for smooth gameplay and a satisfying user experience.

On the other hand, the Gemini 2.5 Pro model also generated a Tetris game that was fully functional. While the visual design may not have been as polished as the GPT-4 Omni version, the game's core functionality was equally impressive, with accurate block placement, rotation, and line clearing mechanics.

Overall, both models were able to successfully complete the Tetris game development prompt, demonstrating their strong coding abilities and understanding of game development principles. This showcases the impressive advancements in language models' capabilities, making them valuable tools for rapid prototyping and development of interactive applications.