DeepSeek V4.1 Flash now available on Databricks Unity Gateway

The DeepSeek V4.1 Flash model has been integrated into Unity Gateway and can be accessed via Foundation Model APIs, supporting text and image multimodal inputs.

Use cases

Imagine you need to process unstructured data and perform fast inference with text and image inputs. With this update, you can:

• Query the DeepSeek V4.1 Flash model using Unity Gateway.
• Submit workloads with text and image inputs.
• Use pay-per-token endpoints directly in your workspace.

How to use

You can send query requests to the model using its model service name in Unity Gateway or via pay-per-token endpoints.

import mlflow.openai

# Example of querying the model using the Unity Gateway service name
response = mlflow.openai.chat(
    model="system.ai.deepseek-v4-1-flash",
    messages=[
        {"role": "user", "content": "Describe this image"}
    ]
)
print(response)

What happens in this code:

1. Imports the MLflow integration library.
2. Sends a chat request using the model service name in Unity Gateway (system.ai.deepseek-v4-1-flash).
3. Returns the generated response from the model for the provided input.

Limitations and requirements

• The model supports text and image inputs.
• Customers are responsible for ensuring compliance with applicable model terms.
• Available in regions that support Foundation Model APIs.