In this video, the creator extensively tests GPT-5’s coding and multimodal capabilities, showcasing its impressive ability to generate complex, interactive programs from single prompts. One of the standout demonstrations is a fully interactive Rubik’s Cube simulation built with 3JS, capable of dynamically adjusting cube sizes up to 20x20x20. The cube not only scrambles and solves visually with smooth animations but also supports camera controls and user interactions like rotating layers. Although some challenges arose with larger cubes, GPT-5 quickly fixed issues through iterative feedback, even benefiting from screenshot inputs to improve its code.
The video also highlights GPT-5’s prowess in creating sophisticated front-end applications. The model successfully built functional clones of Excel and Microsoft Word with features like multiple sheets, formula support, text formatting, image insertion, and undo/redo capabilities—all from single prompts. Other impressive projects include Conway’s Game of Life mapped onto various 3D shapes, a visually enhanced Snake game with particle effects, and a double pendulum physics simulation with adjustable parameters. These examples demonstrate GPT-5’s ability to handle both complex logic and rich user interfaces seamlessly.
Further tests explore GPT-5’s skills in physics simulations and graphics rendering. The model recreated a rotating hexagon with bouncing balls using realistic elastic collisions and developed a cloth simulation with wind and tearable constraints. It also implemented a 2D Navier-Stokes fluid dynamics solver and a minimal ray tracer with interactive controls. While some simulations had minor imperfections, overall, GPT-5 delivered highly detailed and visually appealing results. The video also shows GPT-5’s speed and flexibility, including a “thinking mode” that balances quick answers with deeper reasoning.
Beyond coding, GPT-5’s multimodal abilities were put to the test with image recognition, generation, and analysis. The model accurately identified a location from a blurry photo, generated photorealistic images like raindrops and dragons, and analyzed a children’s book page to find and list 25 errors. It even created new images incorporating all identified errors, showcasing creative image synthesis. The model also handled SVG generation with mixed results and provided thoughtful, responsible advice on sensitive personal questions, demonstrating nuanced understanding and ethical considerations.
Overall, the video presents GPT-5 as a groundbreaking AI with remarkable coding, multimodal, and reasoning capabilities. While some areas still need refinement, especially in complex physics or highly detailed simulations, GPT-5 excels at generating functional, interactive applications and creative content rapidly and with minimal iteration. The creator expresses excitement about future possibilities, including integrating GPT-5 into agentic frameworks for even more powerful applications, and invites viewers to share suggestions and feedback. This comprehensive testing underscores GPT-5’s potential to transform software development and creative workflows.
