Introducing new and updated models to Azure OpenAI Service

Blog Post

Microsoft Foundry Blog

3 MIN READ

From Mercedes-Benz enhancing their in-car experiences to AT&T using these models to empower their employees we’re seeing customers innovate with Azure OpenAI Service like never before.

The gpt-35-turbo and gpt-4 models have been a catalyst for widespread adoption of large language models (LLMs), and we’ve seen tremendous ingenuity from our customers in virtually every industry. We’re excited to announce a few updates to the Azure OpenAI Service:

New versions of gpt-35-turbo and gpt-4 are now generally available.
Price reductions for our most popular models – 75% lower pricing for the state-of-the-art embeddings model and 25% lower pricing on prompt tokens for gpt-35-turbo.

New Models in Azure OpenAI

New versions of both gpt-35-turbo and gpt-4 are now available in Azure OpenAI.

Updates to gpt-35-turbo

There are two new versions of gpt-35-turbo:

gpt-35-turbo (version 0613)
gpt-35-turbo-16k (version 0613)

One of the biggest updates with the 0613 version is enhanced steerability via the system message. This means that the model is better at following the instructions that you outline in the system message which gives you the ability to further customize the behavior of the model or set guardrails to ensure the model is used responsibly.

The 16k version of the model, gpt-35-turbo-16k, offers 4x the context length of the previous gpt-35-turbo model which will help you unlock scenarios requiring longer prompts.

Both new gpt-35-turbo models will now have separate prices for prompt tokens and completion tokens:

	Prompt tokens	Completion tokens
gpt-35-turbo	$0.0015 per 1,000 tokens	$0.002 per 1,000 tokens
gpt-35-turbo-16k	$0.003 per 1,000 tokens	$0.004 per 1,000 tokens

Updates to gpt-4

Both gpt-4 and gpt-4-32k also have a new and improved 0613 model version available that come with the same enhanced steerability. All gpt-4 models are still available by request only. Existing Azure OpenAI customers can apply by filling out this form.

Function calling will be coming soon and will only be available for the new model versions announced today.

Model Lifecycle

To make handling model updates easier, we’ve made some important updates to Azure OpenAI Service. Now when you create a deployment, you can choose whether you want to have the model auto-update to the default version of the model or if you’d like to be pinned to a particular version. You can also go into the Azure AI Studio to update the model version for your deployment at any time.

If you’ve set your deployment to “auto-update to default”, you’ll automatically be upgraded to the new 0613 version of the model in two weeks. Otherwise, your deployment won’t be automatically updated until the model version you’re using is deprecated. You can also choose to set your deployment to never auto-upgrade via our API, although this isn’t recommended as the deployment will start returning 400s after the deprecation date.

To learn more about how model updates work in Azure OpenAI, see https://aka.ms/oai/modelupdates.

Data Privacy in Azure OpenAI Service

When you use Azure services, you are entrusting us with one of your most valuable assets—your data. You trust that the privacy and confidentiality of the data you store and process in Azure services will be protected and that it will be used only in a way that is consistent with your expectations. This means that when you use Azure OpenAI, your data including your prompts and completions, embeddings, training data and finetuned models:

Are never used to improve OpenAI models and is never made available to 3^rd parties.
Are never used to improve any Microsoft or 3^rd party products or services.
Models you finetune are available exclusively for your use.

To learn more about data privacy with Azure OpenAI, see Data, privacy, and security for Azure OpenAI Service. You can also view the list of Azure OpenAI’s compliance certifications.

Getting Started

You can get started today with the new gpt-35-turbo models. If you don’t yet have access to Azure OpenAI Service, you can apply for access by completing the form at https://aka.ms/oai/access.
You can join the GPT-4 waitlist by filling out this form: https://aka.ms/oai/get-gpt4.
We also encourage you to explore and try the service in Azure AI Studio.

Updated Jun 30, 2023

Version 1.0

12 Comments

Why do we not see the version selection field?
It probably should be noted that the 0613 models don't actually work right now.
https://learn.microsoft.com/en-us/answers/questions/1329186/where-did-gpt-3-5-turbo-0613-and-gpt-3-5-turbo-16k
https://learn.microsoft.com/en-us/answers/questions/1329088/gpt-35-turbo-16k-is-not-allowing-)-8k-tokens-(on-a
As of 7/21 they are working.
This is great news, However recently Azure announced that "Some older models in Azure OpenAI Service will be retired on 5 July 2024" . Code-Davinci-002 was one of these models that I was doing experiment with so my questions are:
- Is it possible to keep the retired models and not to go for the replaced model?
- How often do you plan to replace or retire the existing models in the service?
- Is there any other way to check the status and availability of the models, such as an API endpointt?
I am using version 0613, but it seems unstable and takes longer to return results, and the answer is not as good as the 0301 version. I hope you can improve this.
I am getting chatgpt to output string in json format. However the 0613 version seems having more faulty output compared to 0301 version. Anyone feel that the new model capability not as good as the old version?
Hello,
I have a couple of questions regarding the availability of the new versions of GPT, specifically GPT-3.5 and GPT-4, on the EU servers. I would greatly appreciate any insights or updates on these matters.
Firstly, I'm curious to know when we can expect the new versions, GPT-3.5 and GPT-4, to be available on EU servers.
Secondly, I would like to inquire about possible ways to speed up gaining access to GPT-4.
Thank you in advance for any assistance you can provide.
HI DerekLegenzoff , I'm trying to create the new deployment with the latest version of gpt-35-turbo (0613) and getting the below error
Deployment Name: "gpt-35-turbo-deployment": performing CreateOrUpdate: unexpected status 400 with error: InvalidResourceProperties: The specified scale type 'Standard' of account deployment is not supported by the model 'gpt-35-turbo'.
Note: This error is not coming for 0301 version
Update:
I checked the model API and indeed as per the error the new version doesn't support the scale type yet
version: 0613
```
{
"capabilities": {
"fine_tune": false,
"inference": true,
"completion": false,
"embeddings": false,
"scale_types": []
},
"lifecycle_status": "generally-available",
"deprecation": {
"inference": 1705276800
},
"id": "gpt-35-turbo",
"status": "succeeded",
"created_at": 1687132800,
"updated_at": 1687132800,
"object": "model"
},
```
version 0301
```
{
"capabilities": {
"fine_tune": false,
"inference": true,
"completion": true,
"embeddings": false,
"scale_types": [
"standard"
]
},
"lifecycle_status": "generally-available",
"deprecation": {
"inference": 1697328000
},
"id": "gpt-35-turbo",
"status": "succeeded",
"created_at": 1678320000,
"updated_at": 1678320000,
"object": "model"
},
```
May i know the scale_type should support in the future ? or it will not support at all ? Curious to know why one version of the model support where the other version stop supporting it
Great to see lower pricing, that can drive even more usage with customers.
More tokens Is good, but we need functions !! 🙂
This is great to see. Can we make use of the function calling APIs with these models? It seems that the most recent API (2023-06-01) does not support it yet.
UPDATE: I see support is coming soon... It's in this post but I skimmed over it 🙂 Any idea when this is planned?