i think people seriously underestimate how much of intelligence comes down to figuring out what the actual problem is.
some of the hardest problems aren’t difficult because the solutions are complicated. they’re difficult because nobody has correctly identified what needs solving in the first place.
if anthropic has genuinely cracked something here, this could be a far bigger deal than another massive jump in coding benchmarks. give this thing an unsolved scientific problem and let’s see what happens..
holy fucking shit, anthropic might actually be about to do it again.
opus 5.5 is already an absolute monster, and now we’re hearing about another model that could be substantially better.
imagine an agent that understands the problem you’re trying to solve better than you understand how to explain it.
i’m genuinely starting to wonder what the hell their researchers have figured out over there.