Model deprecations

We regularly update our platform with the latest and more advanced open source models. As new models are made available, we routinely retire older models. When we retire a model, it is known as a deprecated model. This document outlines our deprecation process, including communication channels and recommendations for migrating to newer models.

Deprecation process

When a model is being deprecated in SambaNova Cloud, it will never immediately be removed from the platform. It goes through the deprecation process described below.

• This process establishes a seamless transition during model deprecations, ensuring that your applications will continue to operate effectively using the latest advancements in our platform.
• For assistance during the deprecation process, or for any SambaNova Cloud questions, please contact our support team at [e-mail].

Notifications

• Preview models: Models labeled as Preview often graduate to become Production models. Less popular Preview models may be deprecated with minimal notice.
• Production models: Production models will receive a deprecation notice of two to three weeks, depending on usage patterns and the availability of newer versions.

Communication channels

• Email: Users of models scheduled for deprecation will be notified via email. The email will provide details about the deprecation timeline and available alternative models, including recommended replacements.
• Documentation: Model deprecation changes will be reflected in our documentation, including deprecation schedules, migration guides, and recommended replacement models.
• SambaNova Cloud platform: Models scheduled for deprecation will be clearly labeled in the SambaNova Cloud platform to inform users of their upcoming retirement.

Migration options

When a model is scheduled to be deprecated, users have the options described below.

Transition to a newer model

Evaluate and select either the recommended replacement model, or any other model from our serverless offerings, that meets your application’s requirements.

Request a dedicated instance

For critical applications requiring extended support for a deprecated model, users can request a dedicated instance. This instance will be exclusively set up and accessible to you, subject to additional costs and setup time. Please contact our support at [e-mail] to find out more.

Migration steps

To ensure a smooth transition from a deprecated model to a supported alternative, please follow the steps below.

1. Deprecation notice

   Be attentive to email notifications and SambaNova Cloud platform notifications regarding model deprecations.

2. Evaluate alternatives

   Review our documentation to identify and select a suitable replacement model from our offerings.

3. Implement and test

   Integrate your chosen model into your application, conduct necessary testing to ensure compatibility, and deploy your model into production. When migrating to new models, please note that you may need to adjust your prompts to optimize performance.

4. Complete your migration

   To avoid service disruptions, finalize the transition by fully migrating your workload and application to your selected alternative model before the scheduled deprecation date.

Deprecations

All model deprecations are listed below, by date, with the most recent announcements at the top.

April 14, 2025

On March 31, 2025, we notified users of the models listed below about their upcoming retirements. Please transition to the recommended replacement model as soon as possible to avoid any disruptions.

Model ID	Removal date	Recommended replacement
Llama-3.1-Tulu-3-405B	4/14/2025	DeepSeek-V3-0324, Meta-Llama-3.1-405B-Instruct
Llama-3.1-Swallow-70B-Instruct-v0.3	4/14/2025	Llama-3.3-Swallow-70B-Instruct-v0.4
Meta-Llama-3.1-70B-Instruct	4/14/2025	Meta-Llama-3.3-70B-Instruct
Qwen2.5-72B-Instruct	4/14/2025	Meta-Llama-3.3-70B-Instruct, DeepSeek-V3-0324
Qwen2.5-Coder-32B-Instruct	4/14/2025	QwQ-32B
Llama-3.2-90B-Vision-Instruct	4/14/2025	Llama-4-Maverick-17B-128E-Instruct
Llama-3.2-11B-Vision-Instruct	4/14/2025	Llama-4-Maverick-17B-128E-Instruct