Summary
Engineering leadership has requested that all Claude Code / Vertex AI users switch from regional endpoints (e.g. us-east5) to the global endpoint to reduce AI spend. Google charges a premium for tokens served from regional endpoints, and using global allows better cost management across the organization.
Source: ambient-code-go Google Group — forwarded announcement from engineering leadership (2026-04-22).
The official Claude Code installation instructions have been updated to reflect CLOUD_ML_REGION=global.
Current State in This Repo
The platform currently has multiple touchpoints for this setting:
-
operator-config-openshift.yaml — The ConfigMap already sets CLOUD_ML_REGION: "global", but this is overridden at runtime by application code.
-
display_name.go — getAnthropicClient() — Contains explicit fallback logic that overrides global back to us-east5:
region := os.Getenv("CLOUD_ML_REGION")
// Default to us-east5 - claude-haiku-4-5 is not available in global region
if region == "" || region == "global" {
region = "us-east5"
}
-
vertex.go — ValidateVertexConfig() — Reads CLOUD_ML_REGION at startup but does not transform the value.
-
Benchmark scripts (referenced in RHOAIENG-48735) — Default to us-east5 via ${CLOUD_ML_REGION:-us-east5}.
Known Technical Constraint
The override in getAnthropicClient() was added because claude-haiku-4-5 was not available in the global region at the time of implementation. See: https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claude
Action Items
/cc @jeremyeder
Summary
Engineering leadership has requested that all Claude Code / Vertex AI users switch from regional endpoints (e.g.
us-east5) to theglobalendpoint to reduce AI spend. Google charges a premium for tokens served from regional endpoints, and usingglobalallows better cost management across the organization.Source: ambient-code-go Google Group — forwarded announcement from engineering leadership (2026-04-22).
The official Claude Code installation instructions have been updated to reflect
CLOUD_ML_REGION=global.Current State in This Repo
The platform currently has multiple touchpoints for this setting:
operator-config-openshift.yaml— The ConfigMap already setsCLOUD_ML_REGION: "global", but this is overridden at runtime by application code.display_name.go—getAnthropicClient()— Contains explicit fallback logic that overridesglobalback tous-east5:vertex.go—ValidateVertexConfig()— ReadsCLOUD_ML_REGIONat startup but does not transform the value.Benchmark scripts (referenced in RHOAIENG-48735) — Default to
us-east5via${CLOUD_ML_REGION:-us-east5}.Known Technical Constraint
The override in
getAnthropicClient()was added becauseclaude-haiku-4-5was not available in the global region at the time of implementation. See: https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/claudeAction Items
claude-haiku-4-5(and all other Claude models used by the platform) are now available in theglobalregionglobal→us-east5override ingetAnthropicClient()so the ConfigMap value is respectedus-east5defaults in benchmark/test scripts/cc @jeremyeder