4 ms·

I've found it to be marginally worse at some coding tasks than gpt-3.5, but it is all highly specific to the kinds of questions that I'm asking.