Back to docs

Codex multi-region

gpt-5-codex now routes across 8 Azure regions with p50 latency of 1.3 seconds.

Regional footprint

East USWest EuropeSoutheast AsiaAustralia EastJapan EastBrazil SouthNorth EuropeCentral India

Latency profile

1.3sp50
2.1sp95
4.7sp99

How routing works

Each request is dispatched to the nearest healthy Azure region based on client geo-IP. If the primary region is at capacity, the orchestrator falls back to the next-closest region with available GPU quota. All regions share a common model checkpoint, so responses are identical regardless of which region serves the request.