DeepSeek V4.1 Flash supports priority pay-per-token and reserved throughput

The DeepSeek V4.1 Flash model can now be used through Databricks Foundation Model APIs. It supports two main operational modes: priority pay-per-token for latency-sensitive traffic and reserved provisioned throughput for dedicated production workloads. ### Use cases Imagine you need to integrate a fast, efficient model into a customer support application where response time is critical, or into a high-scale production pipeline that requires guaranteed capacity. With this update, you can:- Send inference requests to DeepSeek V4.1 Flash using the Unity Gateway or workspace endpoints.- Configure priority pay-per-token to reduce latency for sensitive traffic.- Ensure dedicated capacity through reserved provisioned throughput for production environments. ### How to use You can interact with the model by sending inference requests via endpoints or through the Unity Gateway. The model reference follows the Unity Catalog naming convention.pythonimport mlflowfrom mlflow.deployments import get_deploy_clientclient = get_deploy_client("databricks")response = client.predict( endpoint="system.ai.deepseek-v4-1-flash", inputs={ "messages": [ {"role": "user", "content": "Explain the concept of Delta tables in a few words."} ] })print(response)What happens in this code:1. Imports the MLflow Deployments client for Databricks.2. Obtains the deployment client pointing to the Databricks environment.3. Sends a prompt to the DeepSeek V4.1 Flash model endpoint via the Unity Gateway.4. Prints the generated response from the model to the console. ### Limitations and requirements- Available via Databricks Foundation Model APIs.- Provisioned throughput is recommended for production workloads to guarantee dedicated capacity.- Customers are responsible for ensuring compliance with applicable model terms.