Aliases vs Stages

Aliases

Stages (Staging, Production, Archived) are a fixed vocabulary baked into MLflow. Real teams often need more nuance: champion, challenger, eu-prod, apac-prod, shadow. Aliases are arbitrary labels you attach to specific versions:

client.set_registered_model_alias('credit-scoring', 'champion', 4)
client.set_registered_model_alias('credit-scoring', 'challenger', 5)

model = mlflow.pyfunc.load_model('models:/credit-scoring@champion')

Aliases are append-only, easy to swap atomically, and survive the move from stages-based to alias-based workflows. MLflow 2.9+ recommends aliases over stages for new projects.

Stages are fixed job titles: Junior, Senior, Lead. Aliases are roles you wear: 'on-call this week', 'tech-lead for project X', 'shadow reviewer'. Roles can be reassigned without inventing a new job title every time the organisation evolves.