Fréchet-type tail distributions enable FTPL to achieve optimal regrets in adversarial bandits and BOBW capability in stochastic bandits.