Upgrade gpt-latest alias to 3-tier failover (gpt-6.1-sol -> gpt-6-sol -> gpt-5.4-mini) #24

Summary

Update the sample's default model deployments and gpt-latest AI Gateway alias to use a 3-tier failover chain across the latest GPT-6 and GPT-5.4-mini models:

1. Priority 0: gpt-6.1-sol (version 2026-09-29, GlobalStandard, capacity 20) — primary flagship model
2. Priority 1: gpt-6-sol (version 2026-09-22, GlobalStandard, capacity 20) — secondary flagship failover in a separate quota pool
3. Priority 2: gpt-5.4-mini (version 2026-03-17, GlobalStandard, capacity 200) — high-capacity safety-net failover (and direct target of the gpt-mini-latest alias)

Note: The exact Foundry catalog model name is gpt-6-sol (not gpt-6.0-sol). Both gpt-6.1-sol (2026-09-29) and gpt-6-sol (2026-09-22) are GA with GlobalStandard SKU across all 6 sample regions (eastus, eastus2, westus, northcentralus, swedencentral, japaneast).

Requirements

1. infra/main.bicep

• Update primary model defaults:
  • modelName: 'gpt-6.1-sol'
  • modelVersion: '2026-09-29'
  • modelDeploymentName: 'gpt-latest' (or keep ARM-safe catalog/deployment naming clean and consistent)
• Add secondary flagship failover model parameters:
  • secondaryModelName: 'gpt-6-sol'
  • secondaryModelVersion: '2026-09-22'
  • secondaryModelDeploymentName: 'gpt-6-sol'
  • secondaryModelCapacity: 20
• Keep mini model defaults:
  • miniModelName: 'gpt-5.4-mini'
  • miniModelVersion: '2026-03-17'
  • miniModelDeploymentName: 'gpt-mini-latest'
  • miniModelCapacity: 200
• Provision all 3 deployments in gatewayModelDeployments and register all 3 in aiGateway's foundryModels.
• Update effectiveModelAliasBackends for the gpt-latest alias (primary-router) to:
  1. modelDeploymentName (gpt-6.1-sol) at priority: 0, weight: 100
  2. secondaryModelDeploymentName (gpt-6-sol) at priority: 1, weight: 100
  3. miniModelDeploymentName (gpt-5.4-mini) at priority: 2, weight: 100
• Keep gpt-mini-latest (mini-router) pointing directly to miniModelDeploymentName at priority: 0, weight: 100.
• Export AI_GATEWAY_INTERNAL_SECONDARY_MODEL_DEPLOYMENT, AI_GATEWAY_INTERNAL_SECONDARY_MODEL_NAME, and AI_GATEWAY_INTERNAL_SECONDARY_MODEL_VERSION outputs alongside existing internal model outputs.

2. Tests (tests/test-ai-gateway-model-registration.sh)

• Update tests/test-ai-gateway-model-registration.sh to assert that effectiveModelAliasBackends in infra/main.bicep configures all three backends (modelDeploymentName at priority 0, secondaryModelDeploymentName at priority 1, and miniModelDeploymentName at priority 2, each with weight 100) and that infra/main.bicep defaults to gpt-6.1-sol (2026-09-29), gpt-6-sol (2026-09-22), and gpt-5.4-mini (2026-03-17).

3. Documentation (IMPLEMENTATION_NOTES.md)

• Update the architecture/model routing narrative and the Regions and model defaults table in IMPLEMENTATION_NOTES.md to document the 3-tier gpt-latest alias (gpt-6.1-sol -> gpt-6-sol -> gpt-5.4-mini) and the fallback guidance.