Feature Request

Is your feature request related to a problem?

When building Grafana dashboards to visualize spend by cloud provider (e.g. Bedrock vs Anthropic direct vs Azure vs OpenAI), the litellm_spend_metric counter lacks the api_provider label, making it impossible to break down cost by provider.

The model label on the spend metric contains the raw model name (e.g. claude-opus-4-6, global.anthropic.claude-opus-4-6-v1, gpt-5.1) with the provider prefix stripped, so you can't reliably determine which provider served the request.

Describe the solution you'd like

Add UserAPIKeyLabelNames.API_PROVIDER to the litellm_spend_metric label list in PrometheusMetricLabels.

The api_provider value is already populated in UserAPIKeyLabelValues from standard_logging_payload["custom_llm_provider"] and already used by other metrics (deployment metrics, batch metrics). It just needs to be added to the spend metric's label list.

This is a one-line change in litellm/types/integrations/prometheus.py.

Describe alternatives you've considered

• Using label_replace() in PromQL to extract the provider from the model label — doesn't work because LiteLLM strips the provider prefix.
• Using deployment metrics (litellm_deployment_success_responses_total) which have api_provider — only gives request count, not spend.
• Creating separate Grafana queries per known model name pattern — brittle and can't distinguish providers that share the same model name (e.g. OpenAI direct vs Azure OpenAI both show as gpt-5.1).

Additional context

The same label could also be useful on litellm_input_tokens_metric and litellm_output_tokens_metric for token-by-provider dashboards.