Acute services for mental health crises are very important to service users and their supporters, and consume a substantial share of mental health resources in many countries. However, acute care is often unpopular and sometimes coercive, and the evidence on which models are best for patient experience and outcomes remains surprisingly limited, in part reflecting challenges in conducting studies with people in crisis. Evidence on best approaches to initial assessment and immediate management is particularly lacking, but some innovative models involving extended assessment, brief interventions, and diversifying settings and strategies for providing support are potentially helpful. Acute wards continue to be central in the intensive treatment phase following a crisis, but new approaches need to be developed, evaluated and implemented to reducing coercion, addressing trauma, diversifying treatments and the inpatient workforce, and making decision‐making and care collaborative. Intensive home treatment services, acute day units, and community crisis services have supporting evidence in diverting some service users from hospital admission: a greater understanding of how best to implement them in a wide range of contexts and what works best for which service users would be valuable. Approaches to crisis management in the voluntary sector are more flexible and informal: such services have potential to complement and provide valuable learning for statutory sector services, especially for groups who tend to be underserved or disengaged. Such approaches often involve staff with personal experience of mental health crises, who have important potential roles in improving quality of acute care across sectors. Large gaps exist in many low‐ and middle‐income countries, fuelled by poor access to quality mental health care. Responses need to build on a foundation of existing community responses and contextually relevant evidence. The necessity of moving outside formal systems in low‐resource settings may lead to wider learning from locally embedded strategies.