For context, the previous error rates were around 0.01% or thereabouts. So, this is a pretty significant advance.

However, this is for a single qubit working in isolation.

For multiple quibits (>1 less than 4) working together the error rate is 0.05% - as you scale beyond that the error rate becomes worse and worse. The errors accumulate and you wind up needing to solve all sorts of other problems: Eliminating crosstalk, increasing coherence, etc.