Baseten adds a second billing path to routing

Hugging Face added Baseten to Inference Providers for conversational and text-generation models, available through its Python and JavaScript clients or an OpenAI-compatible router endpoint. Why it matters: Model routing is not only a model-choice feature. The same provider can be called with a direct Baseten key or routed through Hugging Face, sending charges to different accounts and making cost ownership part of the setup.

Try this: Send one fixed workload through both routes and log the exact model ID, provider, latency, output quality, and billed cost. Do not treat a routed convenience test as a cost comparison until the billing path is explicit.