Understanding Amazon Bedrock model lifecycle
AWS introduces structured 3-state lifecycle with mandatory 6-month legacy period before models sunset.
AWS has introduced a formal, three-stage lifecycle for foundation models (FMs) hosted on its Amazon Bedrock service, providing much-needed predictability for enterprise AI deployments. The lifecycle defines clear states: Active (fully supported), Legacy (with at least 6 months' notice before deprecation), and End-of-Life (EOL), where the model becomes inaccessible. Crucially, AWS mandates that any model launched on Bedrock must remain available for a minimum of 12 months, and once moved to Legacy, it stays there for at least 6 months before reaching EOL. This structured approach is designed to prevent the sudden obsolescence of AI applications.
A key new feature is the 'Public Extended Access' period for models with EOL dates after February 1, 2026. After a minimum of 3 months in Legacy, a model enters this phase, granting active users at least another 3 months of access. During extended access, pricing may be adjusted with advance notice, though customers with private agreements or provisioned throughput are protected. The transition requires proactive action from developers; AWS will not automatically migrate applications, and API calls to EOL models will fail. This framework shifts the responsibility for planning migrations onto development teams but provides the stable timeline necessary to execute them.
- Models have a mandatory 6-month Legacy period before End-of-Life (EOL), giving developers a minimum migration window.
- A new 'Extended Access' phase provides at least 3 additional months of usage for models sunsetting after Feb 2026.
- Pricing during extended access may change, but customers with private agreements or provisioned throughput are shielded from increases.
Why It Matters
Provides enterprises with predictable deprecation timelines, reducing the risk and cost of managing AI application lifecycles on Bedrock.