last couple of releases where I asked it to do a fairly difficult task and have seen how each revision has performed going back to GPT3.
GPT4 almost solved it. I had to do a few manual bugfixes
GPT4o solved it and documented the code well
O1 solved it in a much better way, made suggested improvements, sped up the smal bits of code I provided as inputs, and gave a 5 page explanation of the solution.
Pretty soon even coding is just going to be prompt engineering.