I think these smaller models really struggle with the reasoning aspect of writing decent code, I'm getting pretty nonsensical things out when asking it to fix f

I think these smaller models really struggle with the reasoning aspect of writing decent code, I'm getting pretty nonsensical things out when asking it to fix fizzbuzz (though it made the right fix), like: - Replaced "i % 2 === 0" with "i % 3 === 0", because a number is divisible by 2 only if it's divisible by 3.