Azure AI is committed to delivering the latest AI innovation to help transform your businesses. Our focus with Azure OpenAI Service is to provide a trusted, enterprise-grade platform that enables customers to apply OpenAI’s rapidly advancing models safely and responsibly. It’s been inspiring to see enterprises embracing these models like GPT-4 and GPT 3.5 Turbo, rapidly exploring and prototyping their use cases.
Today, along with OpenAI, we are excited to announce the expansion of the GPT-3.5-Turbo family, new fine-tunable models, and the evolution of the initial GPT-3 models, further enabling our customers to push the boundaries of what's possible.
Introducing new models
In the upcoming weeks, we will do three things:
- Introduce GPT-3.5 Turbo Instruct, the latest addition to the GPT-3.5 Turbo series. This model has been fine-tuned for instruction-following and represents an evolution from the older 'text-[series]-00x' models. As a result, older instruction fine-tuned models, including Text-Davinci-003/002/001, Text-Curie-001, Text-Babbage-001, and Text-Ada-001, will be phased out in favor of GPT-3.5 Turbo Instruct. We anticipate making gpt-35-turbo-instruct generally available at launch.
- Offer a preview of fine-tuning for GPT-3.5-Turbo and GPT-4. We are working on safely enabling fine-tuning for GPT-3.5 Turbo and GPT-4 and expect this feature to be available later this year. This service will initially be available on a limited basis, prioritizing customers who have previously fine-tuned the Davinci series. Fine-tuning of GPT-3.5-Turbo will replace Text and Code-Davinci-002 fine-tunable models.
- Launch previews of new versions of the Ada, Babbage, and Curie models for fine-tuning. Ada-002, Babbage-002, and Curie-002 will replace the older Ada, Babbage, and Curie fine-tunable models.
Retiring older models
In parallel with these announcements, starting today we will retire the older instruct model (‘text-[series]-00x’) and fine-tunable models (Ada, Babbage, Curie and Davinci). They will be unavailable for new customers and they will be retired for existing customers on July 5, 2024.
We will also retire the older embeddings models, including Ada, Babbage, Curie, and Davinci text and code, similarity, and search on July 5, 2024, in favor of text-embedding-ada-002. In January 2023, we launched Text-embedding-ada-002, OpenAI’s state-of-the-art embeddings model, and recently reduced its pricing by 75%. This model has proven to be highly versatile and effective, and has been widely adopted for code or text search and similarity scenarios.
Table 1: Azure OpenAI Service base models and fine-tunable models to be retired and anticipated replacement models
|
Model to be retired
|
Anticipated replacement model
|
|
Base models
|
|
text-ada-001
|
gpt-35-turbo-instruct
|
|
text-babbage-001
|
|
text-curie-001
|
|
text-davinci-001
|
|
text-davinci-002
|
|
text-davinci-003
|
|
code-cushman-001
|
|
code-davinci-002
|
|
Fine-tunable models
|
|
ada
|
babbage-002
|
|
babbage
|
|
curie
|
davinci-002
|
|
davinci
|
Table 2: Azure OpenAI Service embedding models to be retired and replacement model
|
Model to be retired
|
Replacement model
|
|
Embedding models
|
|
code-search-ada-code-001
|
text-embedding-ada-002
|
|
code-search-ada-text-001
|
|
code-search-babbage-code-001
|
|
code-search-babbage-text-001
|
|
text-search-ada-doc-001
|
|
text-search-ada-query-001
|
|
text-search-babbage-doc-001
|
|
text-search-babbage-query-001
|
|
text-search-curie-doc-001
|
|
text-search-curie-query-001
|
|
text-search-davinci-doc-001
|
|
text-search-davinci-query-001
|
|
text-similarity-ada-001
|
|
text-similarity-babbage-001
|
|
text-similarity-curie-001
|
|
text-similarity-davinci-001
|
The majority of our customers are already using GPT-3.5-Turbo, GPT-4, and text-embeddings-ada-002. For customers using the models we are retiring, in the coming weeks we will provide materials to make the transition to existing or replacement models as seamless as possible.
At Azure AI, we remain committed to delivering cutting-edge solutions that empower your businesses. These updates serve as a reflection of this commitment, and we’re excited about the new possibilities they bring. Stay tuned for more updates as we continue to evolve and enhance our service.
This article has been updated with the replacement models.
