ChatGPT-4o, Claude 3 Opus, Gemini 1.0 Ultra, Gemini 1.5 Pro, and ChatGPT-4 responses on the Test of Understanding Graphs in Kinematics (TUG-K), April 2024

The chatbots tested were Google's Gemini 1.0 Ultra, Google's Gemini 1.5 Pro (prompted through Google AI studio using default settings), Anthropic's Claude 3 Opus, OpenAI's ChatGPT-4o (using the latest model GPT-4o) and OpenAI's ChatGPT-4 (using the GPT-4 model). The prompts consisted only of screenshots of the test items. For Gemini 1.0 Ultra, the image was accompanied by the sentence "Answer the question in the image".

The data, collected in April 2024, contains 30 responses from each chatbot to 26 items on the TUG-K survey. For ChatGPT using the GPT-4 model (ChatGPT-4), we provide two separate datasets. One complete dataset (all TUG-K items) without the use "advanced data analysis" plugin, and one with only those six items where the "advanced data analysis" plugin was automaticaly used by the chatbot. These two datasets partially overlap with the ChatGPT-4 dataset published previously (see link below).

This dataset focuses on subscription-based chatbots and is a continuation of a previous dataset that focused on freely available chatbots(10.5281/zenodo.11183803).