It seems that 4o would/could run the code it wrote on sample inputs for test+update cycles, but o1 only pretends to and then reports feelgood results that are a

It seems that 4o would/could run the code it wrote on sample inputs for test+update cycles, but o1 only pretends to and then reports feelgood results that are actually impossible? Is that a known tradeoff that I just missed somewhere in the marketing? Is 4o's appearance of running code also a charade?