After months with Claude and ChatGPT, here is where each one wins
I have been working with Claude and ChatGPT for several months now, using both the free and paid versions, and the more I use them, the clearer it becomes that each one is better at a different type of work.
So, which one do I prefer? The answer depends on whether I want to work with files, write code, produce prose, or give the machine access to documents I care about.
The free versions
When it comes to the free versions, I find Claude more generous. It allows me to download a desktop application, which means that I do not have to work constantly on the browser, and it also allows me to upload and share files more freely.
With ChatGPT, my experience has been more restrictive. I can often upload only one file at a time, and the memory available for the work feels short, so I reach the limit of what I can do much faster. For simple questions this may not matter, but for work that involves several documents or a longer exchange, Claude gives me more room to continue.
For coding, Claude wins
The difference becomes more obvious with the paid versions, particularly when I use them for coding. I am currently migrating feature-engine’s source code to make it compatible with pandas and Polars, using Narwhals to handle both dataframe libraries without sacrificing speed.
I can give Claude that instruction without having to explain every step that follows. It explores the code, makes the changes, runs the tests, investigates the performance, and then provides me with the information I need to decide how to continue. In other words, it does most of the leg work for me.
With ChatGPT, I feel that I need to do more babysitting. I have to tell it what to inspect, what to do next, and sometimes repeat the same instructions, which makes the process less fluid and gives me more of a headache.
That does not mean that I can accept Claude’s code without review. If it were up to Claude, I would have different design patterns for the tests of each feature-engine module, because it can solve each problem separately without considering whether the solution matches the rest of the library. I still need to enforce readability and coherence, and make sure that the same patterns are used across modules, because consistency makes the code easier to understand and maintain.
For writing, ChatGPT wins
When it comes to writing, my preference changes. Claude’s texts tend to sound exactly the same, and they also sound like many of the articles I read online. After using these tools for several months, I can read a piece and identify almost immediately that someone has used Generative AI. When someone submits a pull request, I can also often detect whether it was produced with Claude or with something else.
ChatGPT, on the other hand, learns more readily from my tone of voice and produces more nuanced prose that sounds more human and less artificial. It is not perfect, and I still need to edit it, but I find it easier to move the text towards something that sounds like me.
Here too, review is essential. I need to make sure that the text is relevant and accurate, that it follows a cohesive narrative, and that the machine has not invented information to fill the gaps.
Claude makes me feel safer
There is one more difference, and for me it is an important one. I feel more comfortable working with Claude because it has better guardrails. If I do not give it permission to save, modify, or otherwise alter a file, in most cases it will not do it.
With ChatGPT, I have had the opposite experience. The machine touches, edits, stores, and sometimes deletes content without me having explicitly asked it to do so, which makes me less comfortable giving it access to important documents.
To wrap up
Both tools are useful, but for different reasons. Claude helps me move faster when I code, whereas ChatGPT is better at adapting to my voice when I write. Neither of them, however, removes the need to review the result carefully.
For prose, I need to check relevance, accuracy, and narrative, and make sure that the machine does not invent. For code, I need to enforce patterns of readability and coherence, and also set the priorities, so that a solution does not only work in isolation, but also fits with the rest of the codebase.
It is very rarely that I can ship something that comes out of the machine without review.
What’s your experience?
I hope this information was useful!
Wishing you a successful week ahead - see you next Monday! 👋🏻
Sole
Ready to enhance your skills?
Our courses and books are here to assist you:
