Google's October 8 Gemini API release notes describe the deprecation of Gemini 3.7 Flash and 3.5 Flash, with requests routed to replacement models. The same entry announced an October 23 shutdown for an older Deep Research preview agent.
A working endpoint is not proof of unchanged behavior
Automatic routing can preserve availability while changing output style, reasoning, latency, or downstream parsing behavior. A process that depends on a specific model's quirks may therefore change without a deployment from your own engineering team.
Agent retirements introduce another failure mode: the old identifier may no longer work at all. The changelog distinguishes replacement agent options and migration parameters, so a generic model-upgrade checklist is not sufficient.
Treat release notes as operational input
Maintain an inventory of model identifiers, agent versions, owners, and dependent processes. Record announced retirement dates and set review reminders with enough time for testing. Include sandbox and scheduled jobs; they are easy to miss when only customer-facing applications are inventoried.
Test before switching
Replay representative tasks against the intended replacement, checking business correctness, schema compliance, cost, and unauthorized-action resistance. For research agents, also inspect citation quality and whether the output actually answers the requested question.
Keep the old and new configurations identifiable in your logs. If the provider has already retired an old model, rollback may require a different approved alternative rather than the original identifier. Good maintenance planning makes that distinction before a deadline becomes an incident.
Sources & further reading
Reviewed October 26, 2026. Product availability and regulatory guidance can change. Vendor performance claims are not independent Rovio AI benchmarks.
Found this useful? Share it.