I would consider any system that solves similar problems to be AGI. What I suspect will happen is that this benchmark will saturate long before any such system exists.
When I look at the benchmark questions, they all look like they are exploiting the fact that LLMs suck at composition of subtasks. They might be able to solve each individual problem, but not the combination of them.
Solving that would be a far cry away from AGI.