Comment by stuffoverflow

I guess you are talking about the web interface. I don't think it is a fair way to measure the usefulness of the models because they are clearly instructed to be overly "safe". I think ChatGPT currently has nearly 2000 tokens worth of trash polluting the system prompt where as claude.ai is way too obsessed about things like copyright.

The APIs of all of the models are more permissive and refusals to answer are much more rare.