New feature
GA
Foundation Model APIs
Foundation Model APIs is a Databricks ai / ml capability, introduced December 2023.
Access to state-of-the-art open models from a Databricks-hosted serving endpoint, so you can query a high-quality model without deploying or maintaining one yourself.
- It sells the same models three ways - pay-per-token, priority pay-per-token, and provisioned throughput for workloads that require performance guarantees - so the access mode, not the model, is often the thing you are choosing.
- It reached GA one half at a time: provisioned throughput went generally available in March 2024 and pay-per-token followed five months later.
- The endpoints arrive already existing in your workspace, which is why a brand-new account can query a frontier model before it has created a single table.
Limitations: Pay-per-token endpoints are rate limited per model in input and output tokens per minute and queries per hour, with limits that vary by workspace platform tier and deployment type; provisioned throughput has no TPM limit but is capped at 200 queries per second per workspace with per-model output token limits. Each request is capped at a 4 MB payload and 597 seconds of model execution, and Databricks-hosted models are only available in supported regions.
Open in REbricked →- Category
- AI / ML
- Also known as
- FMAPI, pay-per-token, provisioned throughput
- Verified
- 2026-09-20