[m101]

I have had another experience with Opus 4.8 vs 5.5 which is along the lines of my prior experience of Opus generally having more common sense, whilst GPT is something along the lines of myopic code execution agent:

I was asking them each to review a regulatory pack for planning which regulatory track I needed to follow for my project. The GPT route was to check that the sources were properly referenced and the document structure was accurate and consistent. The Opus approach (the common sense approach, and what I wanted) was to check whether the argument was strong and I had a case.

This type of common sense intelligence difference is something I've seen a lot of times between the two. This doesn't show itself as a difference in a clearly defined coding task however, but there is something there which is qualitatively worse in openai models.