GoonDeepSeek R1 vs OpenAI o1 and o3-mini Models: A Comprehensive Comparison
The AI landscape is rapidly evolving, with new models emerging every week. In this article, we will compare the performance of DeepSeek R1, OpenAI o1, and o3-mini models, as well as Alibaba's new Qwen 2.5 Max model, in three tasks: problem-solving, coding, and web design.
Introduction to the Models
DeepSeek R1 is currently the number one model on the App Store, and we will compare it to other popular models, including o3-mini, o1, and Quen 2.5 Max. We will also test a locally hosted version of DeepSeek R1 with 14 billion parameters.
Introduction to DeepSeek R1 Model
Problem-Solving Task
The first task is a problem-solving challenge, where we provide a piece of code with intentionally introduced mistakes and ask the models to identify and fix the issues. DeepSeek R1 takes 21 seconds to respond and identifies two main issues: a spelling mistake and a CSS error. Quen 2.5 Max responds quickly, identifying not only the spelling mistake and CSS error but also an event delegation issue. OpenAI o1 takes 12 seconds to respond and identifies the spelling mistake and CSS error but not the event delegation issue.
Coding Task
The second task is a coding challenge, where we ask the models to generate code to create a custom mouse cursor when hovering over links. DeepSeek R1 takes 58 seconds to respond and provides a solution that creates a teal circle but does not replace the original cursor. Quen 2.5 Max responds quickly and provides a solution that creates a custom cursor. OpenAI o1 takes! 38 seconds to respond and provides a solution that creates a custom SVG cursor.
Web Design Task
The third task is a web design challenge, where we provide a design with intentionally introduced mistakes and ask the models to identify and recommend fixes. DeepSeek R1 identifies typos, poor visual hierarchy, and poor spacing but does not provide specific recommendations. OpenAI o1 identifies specific issues, such as the title not working, contact information not being clear, and visual hierarchy being incorrect.
DeepSeek R1 Web Design Response
Conclusion
Based on the results, OpenAI o1 performs well across all tasks, followed by Quen 2.5 Max and DeepSeek R1. The locally hosted version of DeepSeek R1 with 14 billion parameters does not perform as well as expected.
Overall, while DeepSeek R1 shows promise, its performance is not as impressive as the hype surrounding it. Quen 2.5 Max performs well in coding tasks, and OpenAI o1 consistently provides smart and relevant responses across all tasks.