Nope, I use a paid version of Chad. Don’t get me wrong, it is a very useful tool and I like it, but it’s way early in its developmental cycle and it absolutely cannot be trusted as a stand-alone resource for data. It doesn’t recognize numbers as data per se (like a calculator would) and that’s why it gets it wrong enough of the time to have to run check-sum balancing against it for validation.
I'm curious what specifically you are calculating and what errors you are finding, that is very different than my experience.
I just learned yesterday that there are effort settings in ChatGPT. I have been using "Auto" which works pretty well, it certainly appears to recognize numbers as data, enough that it can find complex equations in a data sheet, extract them, and correctly calculate the answer. The "Thinking" setting has a charge per use but is even better, and the "Pro" setting is the best and highest cost per use but provides "research grade intelligence".
FWIW, even on the auto setting I have not found any errors in its calculations, even for complex calculus. It's code development capability is amazing. We have been able to delay hiring some additional devs and engineers by using AI, it's pretty impressive.