The fitness cost of antibiotic resistance in the absence of antibiotics is crucial to the success of suspending antibiotics as a strategy to lower resistance. Here we show that after antibiotic treatment the cost of resistance within the complex ecosystem of the mammalian gut is personalized. Using mice as an in vivo model, we find that the fitness effect of the same resistant mutation can be deleterious in a host, but neutral or even beneficial in other hosts. Such antagonistic pleiotropy is shaped by the microbiota, as in germ-free mice resistance is consistently costly across all hosts and in hosts with similar microbiotas the host specific effect of resistance is reduced. An eco-evolutionary model of competition for resources identifies a general mechanism underlying between host variation and predicts that the dynamics of compensatory evolution of resistant bacteria should be host specific, a prediction that was supported by experimental evolution in vivo. The microbiome of each human is close to unique and our results suggest that the short-term costs of resistance and its long-term within-host evolution will also be highly personalized, a finding that may contribute to the observed variable outcome of withdrawing antibiotics to reduce resistance levels.