Debates in homelessness policy often pit Housing First programs, in which subsidized housing is offered unconditionally and immediately, against Treatment First interventions, which combine short-term housing with treatment for issues like substance use. I study a national sample of nearly 300,000 unhoused, mentally ill Veterans who are potentially eligible for the VA's Housing First program (known as HUD-VASH or VA Supportive Housing) or for a Treatment First program (offered by a variety of local contractors). Using an instrumental variables design, I find that enrolling in Housing First reduces three-year mortality by 4.6 percentage points relative to a no-program counterfactual. In contrast, Treatment First has no long-term effects on health. Housing First is also cost-saving, as it causes individuals to substitute away from lengthy inpatient stays.