Your observation is correct—this behavior is expected.

Endpoints with entity_type = FOUNDATION_MODEL_API do not expose health metrics via the /metrics endpoint, which is why you’re getting 404 responses. These endpoints are fully managed, multi-tenant APIs (typically pay-per-token), so infrastructure-level metrics (CPU, memory, etc.) aren’t available through this API.

When /metrics is available

You’ll generally get metrics for infrastructure-backed serving endpoints, such as:

• Custom model serving endpoints
• Feature serving endpoints (feature specs)
• External model endpoints
• Foundation models with provisioned throughput

These support metrics like latency, request count, and error rates.

When /metrics is not available

You’ll typically see 404 in cases like:

• entity_type = FOUNDATION_MODEL_API
• Endpoint is not in a READY state
• Endpoint has no active backing compute

Suggested filtering approach

To avoid unnecessary API calls:

• Skip endpoints where entity_type == FOUNDATION_MODEL_API
• Check that the endpoint is in a READY state before calling /metrics
• Optionally inspect served_entities for more granular filtering

Alternative for FOUNDATION_MODEL_API

For these endpoints, use:

• Usage tracking / billing data (token usage, request logs)
• AI Gateway or system tables for observability