LLMSwaps

Which model each app runs today, and when it last changed

The identifier in your request is the default

A request here names the model identifier it wants, so the platform has no default to publish. The effective one is whatever a customer's code has been saying since somebody wrote it, and the lifecycle state on that model is what tells them when the string needs to change. As of 2026-09-22.

What identifier-based access settles about defaults. Recorded 2026-09-22.
What the page is askedWhat Amazon Bedrock publishesWhat that leaves open
Is a starting tier statedNot applicable, a request names the identifierThe platform never chooses for you
Where the effective default livesIn the customer's own code or configurationUnchanged until somebody edits it
What signals that it should changeThe lifecycle state on that modelReadable through the interface itself
Who performs the changeMigration will not happen automaticallyStated plainly on the lifecycle page

Inclusion rule. Filled when the entry's own public page says which model a new project starts on, or when a single-model menu makes the answer follow from the list having one line. Order. What does not apply first, then where the default actually sits.

1A default that nobody published still exists

The platform does not choose an engine, and yet almost every caller has one it uses without thinking, because a string written eighteen months ago is still in the code. That is a default in every practical sense and it belongs to the customer.

This column records the platform's answer and names the real location of the risk. A production that has never audited which identifiers its own tooling sends is in the same position as one on an unnamed tier.

2The state is the reminder the code cannot give itself

What makes this arrangement better than an unnamed tier is that the model carries a readable state. A caller can check it on a schedule and learn that its long-standing string is now in a legacy window.

Without that, an aging identifier is invisible until it fails. With it, the audit is automatable, which is the difference between a migration that is planned and one that is discovered.

A default nobody published, and a state that can find itAlmost every caller has an identifier it uses without thinking, because a string written months ago is still in the code. What makes this arrangement better than an unnamed tier is that the model carries a readable state, so the aging string is auditable.Your stringAn identifier written once,still in the codeNever revisitedThe real default, owned byyouThe stateActive, Legacy or End-of-Life,on that modelReadable on a scheduleA trigger that does notneed a personThe editMigration will not happenautomaticallyStays with youWork to plan, and asuccessor to chooseWhere the default really sitsWhat each part leaves the customer holding
Fig. 1 Without the state an aging identifier is invisible until it fails; with it, the audit is automatable.

3Nobody moves the string for you

The lifecycle page says migration will not happen automatically, so the edit is the customer's work and so is choosing what to edit it to. No successor is named for any model.

Both halves are recorded in the succession reading. The platform is unusually clear that this is your job, which is more useful than leaving it implied.

4Sources

Read from the model lifecycle page on docs.aws.amazon.com, 2026-09-22. The whole entry is on Amazon Bedrock, and the column it sits in is on Default tier. Same column, other entries: Gemini API, Wan-AI.