from what I'm seeing gemini 4 pro results are either frontier level or weirdly far behind it.
both probably aren't same at the same time.
my current theory is the arena testing.
gemini 4 pro is showing up under "gemini-3.7-flash" label, but it's not routing to 4 pro all the time. so some are just getting actual 3.7 flash results and assuming it's 4 pro.
that would explain why the results being posted are wildly inconsistent.