Still the state of the art, bypassing folly and the swisstables on parallel benchmarks.

And even on single threaded workloads it's about 10x faster than std::unordered_map. My smhasher has benchmark tables.