Databricks schedules DeepSeek V4 Flash (0731) for retirement

The DeepSeek V4 Flash (0731) model in Foundation Model APIs reaches its end of lifecycle and will be completely removed from the environment.

Use cases

Imagine you have applications and endpoints integrated with this specific version of DeepSeek. With support ending, these pipelines will stop working unless migrated.

• Schedule the migration to the recommended new version.
• Map which users and endpoints still rely on the legacy model.
• Prevent production interruptions by updating API calls in time.

How to use

Identify which routines and endpoints in your organization still depend on the deprecated model by running a SQL query against the system catalog.

SELECT
   eu.requester,
   se.endpoint_name,
   se.entity_name,
   COUNT(*) AS request_count,
   SUM(eu.input_token_count) AS total_input_tokens,
   SUM(eu.output_token_count) AS total_output_tokens,
   MIN(eu.request_time) AS first_request,
   MAX(eu.request_time) AS last_request
 FROM system.serving.endpoint_usage eu
 JOIN system.serving.served_entities se
   ON eu.served_entity_id = se.served_entity_id
 WHERE LOWER(se.entity_name) LIKE '%deepseek-v4-flash%'
 GROUP BY eu.requester, se.endpoint_name, se.entity_name
 ORDER BY request_count DESC

What happens in this code:

1. Joins endpoint usage telemetry with served entities from the system catalog.
2. Filters by the name of the model scheduled for retirement.
3. Groups by requester and endpoint to aggregate request volume and token consumption.
4. Orders the results to highlight the most active endpoints requiring urgent migration.

Limitations and requirements

DeepSeek V4 Flash (0731) on pay-per-token is scheduled for retirement on November 5, 2026. The recommended replacement version is DeepSeek V4.1 Flash.