Applications that use the stable model name of the base GPT-3 model (ada, babbage, curie, davinci) will be automatically upgraded to the new model listed above on January 4, 2024. You will also be able to access the new model in the coming weeks for initial testing by specifying the following model name in your API call: babbage-002, davinci-002.
Other older completion models ( text-davinci-003), you must manually upgrade your integration by January 4, 2024. gpt-3.5-turbo-instruct Included in the “model” parameter of the API request. gpt-3.5-turbo-instruct is an InstructGPT-style model, trained similarly to: text-davinci-003. This new model is a drop-in replacement for the Completions API and will be available for initial testing in the coming weeks.
Developers who wish to continue using the fine-tuned model after January 4, 2024 will need to fine-tune their replacement on top of the new base GPT-3 model (babbage-002, davinci-002), or a new model (gpt-3.5-turbo, gpt-4). When this feature becomes available later this year, users who previously tweaked older models will be given priority access to his GPT-3.5 Turbo and GPT-4 tweaks. We recognize that moving away from models that are fine-tuned based on proprietary data is difficult. To help make this transition as smooth as possible, we will provide support to those who have previously tweaked their models.
In the coming weeks, we will be contacting developers who have recently used these older models and will provide more information once we are ready for initial testing of the new completed models.