Exactly. I measured vector vs an efficient linear probing hash map very recently and the cutoff was single digit. Even against unordered_map or plain std::map the cutoff was surprisingly low (although in this case I would trust a synthetic benchmark significantly less).