Model Retirement Runbooks is an operational concern, not merely a setting to turn on. The correct approach separates provider-documented behavior from decisions the application must enforce itself, with explicit evidence, budgets and recovery paths.
What the dated release actually establishes
DeepSeek documents changed alias targets and Anthropic publishes model deprecation and tentative retirement tables. Those notices illustrate why a migration plan must distinguish discontinued IDs from older IDs still serving through redirects.
Where it is useful in a real workflow
Start a release register with provider, old ID, target ID, planned shutdown, earliest affected user, region and use-case owner. Run shadow tests on a redacted test set, then switch low-risk traffic gradually.
Limitations, caveats and failure modes
A planned retirement is not the same as actual disablement; a temporary alias redirect is not model identity preservation. A transition may change token counts, refusal policies and tool-call schemas.
Alternatives and a decision rule
Have a rollback decision, but recognize that the original endpoint may already be unavailable when a rollback is attempted. Preserve output fixtures and a provider-independent failure mode.
Version, region and tier verification
No company or project API keys are embedded in documentation. Each dated provider lifecycle entry needs independent verification.
Minimal production checklist
Use a purpose-built test fixture before treating this as production architecture. Define a permitted input, an expected output schema, an error policy and one measurable performance objective. Log a request correlation ID, model and endpoint identity, duration, token counters when exposed, and validation outcome. Simulate one expired credential, one rate-limit response, one malformed output, one timeout and one human cancellation. A successful demonstration is not the same as a fully recoverable service.
How to audit the outcome
A second reviewer should be able to replay the process without relying on unstated defaults. Record the API and SDK versions, selected region, any caching behavior, response codes and the exact decision to retry, reject, redact or escalate. Keep test data within its allowed security boundary. Account for all failed calls and cleanup operations when calculating latency and cost; otherwise the apparent productivity gains can be an accounting artifact. Compare the final operational results with an explicit baseline, and state clearly when a provider document supports a capability but no independent test has been run.
Sources & Further Reading
- DeepSeek API changelog (retrieved from first-party public documentation on 2026-10-25; this is not a deployment test)
- platform.claude.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.