Azure AI models in qatarcentral now show aiModels.openai.gpt-5-4-mini.2026-03-17 as unavailable after 58 available days in a 67-day history. This is a deprecation candidate rather than a confirmed retirement, but it does reduce one model/version option in the Middle East from 30/64 covered regions to 29/64.
Operationally, that means teams using Qatar Central should re-check deployment targets, fallback model/version choices, and any residency or latency assumptions tied to that region. The monitor records 13.4% unavailable time for this feature, with no prior disappearances, so this is the first observed drop rather than a recurring churn pattern.
At the same time, the model-router feature set expanded into new regions. In South America, brazilsouth is the first observed region for aiModels.openai.model-router.2025-05-19, .2025-08-07, and .2025-11-18. That broadens placement choices for routing-driven workloads where latency and data residency matter.
North America and Europe also picked up the same router versions in eastus, northcentralus, and polandcentral. The rollout lifts feature coverage from 5/64 to 9/64 regions, with same-day additions across brazilsouth, eastus, northcentralus, and polandcentral.
These router signals matter for capacity planning and model selection because they add regional options for inference routing without changing the feature note’s core use case: evaluating model capabilities, deployment options, latency, and data residency.