Deprecation

Deprecation refers to the process of retiring older models or endpoints in favor of hosting better models with better capabilities for you to leverage. When we announce that a model or endpoint is being deprecated, we will provide a shutdown date on which the model or endpoint will no longer be accessible. As such, your applications relying on Groq may need occasional updates to continue working.

Once a model is announced as deprecated, make sure to migrate usage to a recommended replacement before the shutdown date to avoid failing requests. All API deprecations along with recommended replacements are listed below.

Deprecation History

December 20, 2024: Llama 3.1 70B

On December 6, 2024, in partnership with Meta, we released llama-3.3-70b-versatile and notified users that we would deprecate llama-3.1-70b-versatile in favor of hosting Llama 3.3 with significant quality improvements for a better experience.

To facilitate a smooth transition, we will maintain the current llama-3.1-70b-versatile until December 20, 2024. At that time, requests to this model ID will automatically upgrade to llama-3.3-70b-versatile. While this new model delivers improved quality, it may produce different responses than its predecessor. If you need to maintain exact response patterns, we recommend migrating to explicitly using llama-3.3-70b-versatile before December 20, 2024, for testing.

December 18, 2024: Gemma 7B

On December 11, 2024, we emailed all Gemma 7B users that we would deprecate it in favor of keeping the Gemma 9B model as it offers better performance.

November 25, 2024: Llama 3.2 90B Text Preview

In November 2024, we emailed all Llama 3.2 90B Text Preview users that we would deprecate it in favor of hosting the Llama 3.2 90B Vision Preview model for vision capabilities.

October 18, 2024: LLaVA 1.5 7B and Llama 3.2 11B Text Preview

In September 2024, we made Meta's Llama 3.2 vision models available on GroqCloud and emailed all LLaVA 1.5 7B and Llama 3.2 11B Text Preview users that we would deprecate it in favor of hosting Llama 3.2 11B Vision for better performance and more robust vision capabilities.