AI models maintenance policy
This page describes the model maintenance policy for the Foundation Model APIs pay-per-token, Foundation Model APIs provisioned throughput, and Batch Inference with ai_query offerings.
To continue supporting the most state-of-the-art models, Databricks manages models through a lifecycle that progresses from legacy status to deprecation to eventual retirement.
- Legacy: When a model is marked legacy, it is no longer recommended for new workloads, but remains available in workspaces with existing usage of the model. Workspaces that are not using the model when it enters legacy status no longer have access to the model. Customers using such a model should evaluate newer models and consider migration.
- Deprecation: When a model is deprecated, a retirement date of 30 days or 90 days in the future is announced. It is not recommended for new workloads, but remains available in workspaces with existing usage of the model. Workspaces that are not using the model at the time of deprecation do not have access to the model. Customers using such a model should evaluate newer models and plan to migrate their workloads before the retirement date.
- Retirement: When a model is retired, it is no longer accessible and support for the model is fully discontinued. Any workload using the model no longer works.
Model lifecycle policy
The following section summarizes the model lifecycle policy. See Deprecated models for a list of deprecated models and retirement dates.
Many factors are considered when determining the retirement timeline, including model popularity, suitable replacement model, and amount of production usage. Databricks announces retirement dates for models when they are deprecated, following the notification timelines outlined below.
Foundation Model APIs
The following table summarizes the model lifecycle policy for Foundation Model APIs products: pay-per-token, provisioned throughput, and ai_query (batch inference).
Legacy | Deprecation notification and transition to retirement | On the retirement date |
|---|---|---|
Databricks takes the following steps to inform customers that a model is marked legacy:
When a model is marked legacy:
| Databricks takes the following steps to notify customers about a model deprecation:
When a model is deprecated, Databricks announces a retirement date 30 days or 90 days in the future. During this transition period:
| The model is no longer available for use and is removed from the product. Any existing workloads using the model stop working. Applicable documentation is updated to indicate that the model is no longer available, and recommends use of a replacement model. |
Partner model retirement policy
Partner models are models that third-party partners — specifically OpenAI, Anthropic, and Google — provide through Foundation Model APIs. For these partner models, Databricks generally follows the same deprecation timelines and policies described above.
However, partners might provide retirement dates shorter than the one-month transition period that Databricks publishes. In these cases, Databricks attempts to bridge the gap by temporarily redirecting models to a similar version, so customers receive the full transition time.
For example, if a partner model retirement is announced with two weeks' lead time instead of one month, Databricks redirects the model for an additional two weeks to prevent immediate breakage and allow time for migration. Queries fail at the end of the full one-month period.
This redirection can only occur if the replacement model has the same price and is backwards compatible. The replacement model is usually an incremental model version, like 3.0 versus 3.1.
Deprecated and retired models
The following sections list models that are deprecated (no longer recommended for new workloads) or retired (end-of-life and no longer available). Retirement dates for deprecated models are announced at least one month in advance.
Foundation Model APIs retirements
The following table shows model retirements, their retirement dates, and recommended replacement models to use for Foundation Model APIs pay-per-token and provisioned throughput serving workloads. Databricks recommends that you migrate your applications to use replacement models before the indicated retirement date.
Retirement date | Recommended replacement model | |
|---|---|---|
Anthropic Claude Sonnet 4 | Pay-per-token: October 9, 2026 | Claude Sonnet 4.6 |
OpenAI GPT-5.1 Codex Max | Pay-per-token: July 16, 2026 | OpenAI GPT-5.5 |
OpenAI GPT-5.1 Codex Mini | Pay-per-token: July 16, 2026 | OpenAI GPT-5.4 Codex Mini |
OpenAI GPT-5.2 Codex | Pay-per-token: July 16, 2026 | OpenAI GPT-5.5 |
Anthropic Claude 3.7 Sonnet | Pay-per-token: April 12, 2026 | Use the latest Claude Sonnet model |
Gemini 2.5 Flash | Pay-per-token: October 2, 2026 Provisioned throughput: October 2, 2026 | Gemini 3.1 Pro or Gemini 3.5 Flash |
Gemini 2.5 Pro | Provisioned throughput: October 2, 2026 | Gemini 3.1 Pro or Gemini 3.5 Flash |
Gemini 3 Pro | Provisioned throughput: March 26, 2026 | Gemini 3.1 Pro. To allow more time for migration, between March 26, 2026 and June 7, 2026, API calls to Gemini 3 Pro will be temporarily redirected to Gemini 3.1 Pro. The pricing for both models is identical. |
Open model | Retirement date | Recommended replacement model |
|---|---|---|
DeepSeek V4 Flash (0731) | Pay-per-token: November 5, 2026 | DeepSeek V4.1 Flash |
Thinking Machine Labs Inkling | Pay-per-token: October 30, 2026 | GLM 5.3 or Kimi K3 |
DeepSeek V4 Pro (0813) | Pay-per-token: October 30, 2026 | DeepSeek V4.1 Flash |
Kimi K2.7 | Pay-per-token: October 30, 2026 | Kimi K3 |
Meta Llama 3.1 405B | Pay-per-token: February 15, 2026 Provisioned throughput: May 15, 2026 | OpenAI GPT OSS 120B |
DBRX / DBRX Instruct | Pay-per-token: April 30, 2025 Provisioned throughput: December 19, 2025 | Pay-per-token: Meta-Llama-4-Maverick Provisioned throughput: Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Mixtral 8x7B / Mixtral-8x7B Instruct | Pay-per-token: April 30, 2025 Provisioned throughput: February 27, 2026 | Pay-per-token: Meta-Llama-4-Maverick Provisioned throughput: Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Meta Llama 3 (70B) | Pay-per-token: July 23, 2024 (Meta-Llama-3-70B-Instruct); December 11, 2024 (Meta-Llama-3.1-70B-Instruct) Provisioned throughput: February 27, 2026 | Pay-per-token: Meta-Llama-4-Maverick Provisioned throughput: Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Meta Llama 3 8B | Provisioned throughput: February 27, 2026 | Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Meta Llama 2 70B / Meta-Llama-2-70B-Chat | Pay-per-token: October 30, 2024 Provisioned throughput: February 27, 2026 | Pay-per-token: Meta-Llama-4-Maverick Provisioned throughput: Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Meta Llama 2 13B | Provisioned throughput: February 27, 2026 | Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Meta Llama 2 7B | Provisioned throughput: February 27, 2026 | Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
Mistral 7B | Provisioned throughput: February 27, 2026 | Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
MPT 30B / MPT 30B Instruct | Pay-per-token: August 30, 2024 Provisioned throughput: December 19, 2025 | Pay-per-token: Meta-Llama-4-Maverick Provisioned throughput: Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
MPT 7B / MPT 7B Instruct | Pay-per-token: August 30, 2024 Provisioned throughput: December 19, 2025 | Pay-per-token: Meta-Llama-4-Maverick Provisioned throughput: Comparable model on the same offering, like Llama 3.2, 3.3, or 4 model of similar size. |
If you require long-term support for a specific model version, Databricks recommends using Foundation Model APIs provisioned throughput for your serving workloads.
Find workloads that use retired models
Use the following query to find workloads that are using deprecated models and identify their owners.
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 '%<retired-model-name>%'
GROUP BY eu.requester, se.endpoint_name, se.entity_name
ORDER BY request_count DESC