ChatGPT is my go-to for most things, but sometimes, it just doesn’t cut it. DeepSeek is proving itself to be a powerful model that can directly compete with ChatGPT—and even outmatch it in several key tasks.
4
Mathematical Problem-Solving
AI chatbots like DeepSeek and ChatGPT are popular platforms where people go to get assistance and solve math problems. DeepSeek uses its R1 model for reasoning tasks, while ChatGPT provides OpenAI’s newer o3-mini (low/medium) for its free tier users and o3-mini (high) for its Plus tier at a maximum of 50 prompts per day.
After testing dozens of hard GMAT (Graduate Management Admission Test) problems on both DeepSeek and ChatGPT (as a free user), they both provided correct answers to all the problems.
Though this test wasn’t all that extensive, I would say both models are likely good enough to solve common math problems, and you’d probably be hard-pressed to find one that both models couldn’t answer.
This said, I still prefer using DeepSeek over ChatGPT as it scored higher in both AIME Math 2024 and Codeforces benchmarks. DeepSeek’s chain-of-thought also gives more insights into how the problems were solved, allowing me to better understand and educate myself on how to tackle similar problems in the future.
If you are a ChatGPT Plus user, DeepSeek may still be the better option as it won’t use your o3-mini (high) prompt allowance, it provides better chain-of-thoughts, and would probably still solve your mathematical problems unless it’s a theoretical one.
3
Code Generation and Debugging
Coding and debugging are other popular applications where DeepSeek and ChatGPT are used. As stated earlier, DeepSeek’s R1 model scores higher than OpenAI’s o3-mini (low/medium) models in the Codeforces benchmark, which is already a good reason to use DeepSeek over ChatGPT.
To see how that translates into actual usage, I’ve prompted both chatbots to write a snake game with HTML5, CSS, and JavaScript. After a few more prompts to deal with errors, I did eventually get both chatbots to produce a working snake game.
What I’ve noticed is that DeepSeek requires slightly less prompting to fix the issues. But that didn’t really prove much as I got ChatGPT’s snake game to work flawlessly after two more prompts. However, what did make a difference was that DeepSeeks’s snake game was more polished and had more features than the one that came from ChatGPT.
So, although both AI models scored pretty even in the benchmarks, DeepSeek’s R1 seems to provide a bit more hand-holding in terms of what it thinks a user might want the code to be like.
Some may like ChatGPT more for that reason, but I’d argue that most people generating code with chatbots are likely students and junior engineers looking for assistance. Thus, providing extra features that you’d typically find in similar pieces of code would be a plus and a good reason to continue using DeepSeek.
2
Data Analysis
DeepSeek’s strength in data analysis comes from its use of a Mixture of Experts (MoE) model architecture. This design enables the model to dynamically allocate specific subsets of its parameters (”experts”) to different tasks, optimizing computational resources and enhancing processing efficiency. Such a structure allows DeepSeek to effectively handle both structured and unstructured data.
In this example, I gave both DeepSeek and ChatGPT a seed file that I use to populate a database for testing backends. I then prompted both chatbots to analyze potential trends based on the file I’ve provided. DeepSeek was able to provide me with valuable insights such as price distribution, inventory level, peak and recent activity, and collection popularity, etc.
In contrast, ChatGPT seemed to be more concerned about the quality of information in the file. It then proceeds to give advice on how to do data analysis instead of actually doing the execution. I even tried a few times to prompt it to see trends in price distribution, inventory level, peak activity, and recent activity (the trends DeepSeek already found), but was continually given instructions instead.
This is where finding the right AI tool for the job really shines. Though ChatGPT’s free o3-mini models may be better at conversational and creative work, DeepSeek’s R1 model has been specialized for more analytical workloads.
1
Structured Data Processing
DeepSeek’s efficiency in handling structured data sets it apart from general AI models like ChatGPT. Structured data, such as JSON files, XML, and database entries, requires precise parsing and interpretation.
Though it scored lower in GPQA (Graduate-Level Google-Proof Q&A) benchmarks, it doesn’t really matter as much as DeepSeek’s ability to do logic and reasoning, especially when working with structured data.
In this test, I gave both chatbots a misconfigured database for them to process and organize properly.
DeepSeek provided me with tabulated results which were exactly what the database was supposed to look like, while ChatGPT seemed to struggle and gave me only the category section of the database and forgot everything else.
Although I’m confident that with a few more prompts, I could eventually make ChatGPT work on formatting and organizing a small database, this test shows that DeepSeek understood the task on the first try, saving me time and effort in trying to process structured data. Overall, DeepSeek’s deep chain of thoughts and MoE architecture make it stand out from all the ChatGPT alternatives available.
DeepSeek’s strengths lie in its reasoning and ability to handle complex tasks with high accuracy. Though it may not be the best for creative work and general conversations, its advanced mathematical capabilities, superior coding assistance, efficient data analysis, and structured data processing make it my go-to AI tool for these specialized tasks.
