Even with this new model, at the bottom of the page I see something like "Claude can't yet run the code it generates." But if I ask it to run the code it has ge

Even with this new model, at the bottom of the page I see something like "Claude can't yet run the code it generates." But if I ask it to run the code it has generated for me, it confidently generates output that looks like the output of that code should look, with the wrong numbers (off by a factor of about a million in the case of my toy question). When I tell it it's off by a factor of a million it regenerates the output, and is wrong again.