A development group is crafting an AI chatbot proof-of-concept. Their key needs include connecting to an established large language model, avoiding infrastructure setup or upkeep. With expected average usage and prioritizing quick initial deployment over guaranteed performance, what's the best Foundation Model API strategy?
Answer & explanation
Correct answer: B
WHY B is correct: Pay-per-token Foundation Model APIs provide pre-configured, immediately accessible endpoints (no deployment required) and are explicitly documented as 'recommended for getting started.' They support production workloads and moderate traffic. The developer simply calls the pre-existing endpoint with an API token and begins querying.
WHY NOT A: Provisioned throughput is recommended for production workloads requiring performance guarantees, HIPAA compliance, or fine-tuned models. It requires configuring a provisioned throughput endpoint — additional setup that is unnecessary for a quick proof-of-concept.
WHY NOT C: External model serving is for routing to models hosted outside Databricks (e.g., OpenAI, Anthropic via API). The question does not indicate a requirement to use third-party non-Databricks-hosted models.
WHY NOT D: Custom PyFunc model serving is required when the team needs custom inference logic wrapping a model. For a standard chat application using a foundation model, no custom PyFunc wrapper is needed — the Foundation Model API provides a standard OpenAI-compatible interface directly.