I wanted to test some important features that have not been done extensively before, and that can be really useful, rather than relying on conclusions drawn from companies’ own public statements, which can be accessed from anywhere, and I hope these will be very useful to you when making your decision. Let’s push the boundaries of Anthropic’s most advanced model, Claude 3.5 Sonnet, and OpenAI’s most advanced model, GPT 4o, and make a comparison.
Firstly, let’s look at the graphics rendering features. I give a text to GPT 4o and ask it to create a graph for me from the information in the text. Text: ‘I have some data for the techcrunch.com website, create a graph for me using this data. Visit Data, April: 11.74 million, May: 10.54 million, June: 9.70 million’. As you can see in the images below, the result was really satisfactory. It even created a bar chart for me as a second version of the graph. The data I gave were shown in the correct positions on the graphs without error.
Now let us direct the same question to Claude 3.5 Sonnet. The result was negative, as can be seen below. Claude said that although he could understand what I wrote, he could not make a graph. He also told me which data to use and how to use it to create a graph.
In our next test, we asked in which languages the artificial intelligence models speak with the sentence ‘in which languages can you speak?’. The answers we received from GPT 4o and Claude 3.5 were similar and quite predictive. both primarily said that it supported a large number of languages and gave a list, but also stated that it could speak in a large number of languages that were not listed. This was a valid result for both models.
The next test concerns voice interaction. While you can talk to GPT 4o and ask it questions, Claude 3.5 is text-only and cannot input or output any audio. It is safe to say that GPT 4o is much more advanced in this respect.
For the next test, I have a small sample data set and I want to analyse this data set. First of all, let’s send GPT 4o our data in the table below and ask it to analyse it and create a model for us. I send the data with an excel file and I use the following text ‘I have some data in the file I will send and I want you to analyse this data and create a model for me’.
The answer is quite good, it visualised it with Matplotlib Chart and looked at the correlation between each variable, analysed the data with regression and as a result gave me the mathematical model of the regression model so that I can use it in predictions.
Mathematical Output of the Model: House Price= −50,637.63 + 155.99 × House Size (sqft) + 35,419.85 × Number of Rooms + 678.04 × House Age (years)
Given this mathematical model, I can now make predictions. For example, to estimate the house price of a new house, if I have the size, number of rooms and age of the house, I can estimate the price by placing them in this model and making calculations. From simple analyses to complex analyses, GPT 4o makes our jobs much easier, doesn’t it?
Now let’s try the same thing in Claude 3.5 and see the results. Similarly, I wanted to upload my excel file and have it analyzed, but it didn’t accept my file and told me that it couldn’t display the data from the file and that I had to write the data, so I wrote the data manually and ran it. But convrting the data to plain text is a very difficult and tedious task. Because when you have hundreds or thousands of lines of data, it will be impossible to do this.
I was able to enter it manually, albeit with some effort, because my data size was small, but I think those who deal with large data other than making samples cannot use this feature because they cannot attach files. Anyway, the results and evaluations were similar to those of GPT 4o, but the statistics did not produce outputs. Claude did some calculations and explanations, but the mathematical model was also not available even though I asked for it. I have to say that I am at the limits of his capacity to understand and produce based on his output.
I could do a lot more of the above comparison scenarios, but I think the outcome for me is obvious. I think GPT 4o is an AI model that is way, way ahead of Claude 3.5. Maybe in future updates the Claude model will be improved and do better, but for now the conclusion is clear.
Today, it is not enough for AI models to communicate only through text, there are now many AI models that do this. In order to respond to the needs of users, we need models that maximize interaction with the user, receive files, produce visuals and graphics, and speak.