@@ -9,7 +9,7 @@ ms.custom: build-2023, build-2023-dataai, devx-track-python, devx-track-dotnet,
 ms.topic: quickstart
 author: mrbullwinkle
 ms.author: mbullwin
-ms.date: 08/31/2023
+ms.date: 09/20/2024
 zone_pivot_groups: openai-quickstart-new
 recommendations: false
 ---

@@ -6,7 +6,7 @@ author: mrbullwinkle
 ms.author: mbullwin
 ms.service: azure-ai-openai
 ms.topic: conceptual 
-ms.date: 04/30/2024
+ms.date: 09/20/2024
 ms.custom: template-concept
 manager: nitinme
 ---

@@ -3,7 +3,7 @@ title: Azure OpenAI Service getting started with customizing a large language mo
 titleSuffix: Azure OpenAI Service
 description: Learn more about the concepts behind customizing an LLM with Azure OpenAI.
 ms.topic: conceptual
-ms.date: 03/26/2024
+ms.date: 09/20/2024
 ms.service: azure-ai-openai
 manager: nitinme
 author: mrbullwinkle
@@ -76,9 +76,9 @@ Fine-tuning requires the use of high-quality training data, in a [special exampl
 
 ### Illustrative use case
 
-An IT department has been using GPT-4 to convert natural language queries to SQL, but they have found that the responses are not always reliably grounded in their schema, and the cost is prohibitively high.
+An IT department has been using GPT-4o to convert natural language queries to SQL, but they have found that the responses are not always reliably grounded in their schema, and the cost is prohibitively high.
 
-They fine-tune GPT-3.5-Turbo with hundreds of requests and correct responses and produce a model that performs better than the base model with lower costs and latency.
+They fine-tune GPT-4o mini with hundreds of requests and correct responses and produce a model that performs better than the base model with lower costs and latency.
 
 ### Things to consider
 
@@ -90,13 +90,13 @@ They fine-tune GPT-3.5-Turbo with hundreds of requests and correct responses and
 
 - Fine-tuning costs:
 
-  - Fine-tuning can reduce costs across two dimensions: (1) by using fewer tokens depending on the task (2) by using a smaller model (for example GPT 3.5 Turbo can potentially be fine-tuned to achieve the same quality of GPT-4 on a particular task).
+  - Fine-tuning can reduce costs across two dimensions: (1) by using fewer tokens depending on the task (2) by using a smaller model (for example GPT-4o mini can potentially be fine-tuned to achieve the same quality of GPT-4o on a particular task).
 
   - Fine-tuning has upfront costs for training the model. And additional hourly costs for hosting the custom model once it's deployed.
 
 ### Getting started
 
 - [When to use Azure OpenAI fine-tuning](./fine-tuning-considerations.md)
 - [Customize a model with fine-tuning](../how-to/fine-tuning.md)
-- [Azure OpenAI GPT 3.5 Turbo fine-tuning tutorial](../tutorials/fine-tune.md)
+- [Azure OpenAI GPT-4o Turbo fine-tuning tutorial](../tutorials/fine-tune.md)
 - [To fine-tune or not to fine-tune? (Video)](https://www.youtube.com/watch?v=0Jo-z-MFxJs)
\ No newline at end of file

@@ -4,7 +4,7 @@ titleSuffix: Azure OpenAI
 description: Learn about model versions in Azure OpenAI. 
 ms.service: azure-ai-openai
 ms.topic: conceptual 
-ms.date: 10/30/2023
+ms.date: 09/20/2024
 manager: nitinme
 author: mrbullwinkle #ChrisHMSFT
 ms.author: mbullwin #chrhoder
@@ -15,15 +15,11 @@ recommendations: false
 
 Azure OpenAI Service is committed to providing the best generative AI models for customers. As part of this commitment, Azure OpenAI Service regularly releases new model versions to incorporate the latest features and improvements from OpenAI.
 
-In particular, the GPT-3.5 Turbo and GPT-4 models see regular updates with new features.  For example, versions 0613 of GPT-3.5 Turbo and GPT-4 introduced function calling.  Function calling is a popular feature that allows the model to create structured outputs that can be used to call external tools.
-
 ## How model versions work
 
 We want to make it easy for customers to stay up to date as models improve.  Customers can choose to start with a particular version and to automatically update as new versions are released.
 
-When a customer deploys GPT-3.5-Turbo and GPT-4 on Azure OpenAI Service, the standard behavior is to deploy the current default version – for example, GPT-4 version 0314.  When the default version changes to say GPT-4 version 0613, the deployment is automatically updated to version 0613 so that customer deployments feature the latest capabilities of the model.
-
-Customers can also deploy a specific version like GPT-4 0613 and choose an update policy, which can include the following options:
+When you deploy a model you can choose an update policy, which can include the following options:
 
 * Deployments set to **Auto-update to default** automatically update to use the new default version.
 * Deployments set to **Upgrade when expired** automatically update when its current version is retired.

@@ -4,7 +4,7 @@ titleSuffix: Azure OpenAI Service
 description: Learn about how red teaming and adversarial testing are an essential practice in the responsible development of systems and features using large language models (LLMs)
 ms.service: azure-ai-openai
 ms.topic: conceptual
-ms.date: 11/03/2023
+ms.date: 09/20/2023
 manager: nitinme
 author: mrbullwinkle
 ms.author: mbullwin

@@ -4,7 +4,7 @@ titleSuffix: Azure OpenAI Service
 description: Learn about how to construct system messages also know as metaprompts to guide an AI system's behavior.
 ms.service: azure-ai-openai
 ms.topic: conceptual
-ms.date: 03/26/2024
+ms.date: 09/20/2024
 ms.custom:
   - ignite-2023
 manager: nitinme
@@ -69,7 +69,7 @@ Here are some examples of lines you can include:
 
 ## Provide examples to demonstrate the intended behavior of the model
 
-When using the system message to demonstrate the intended behavior of the model in your scenario, it is helpful to provide specific examples. When providing examples, consider the following:
+When using the system message to demonstrate the intended behavior of the model in your scenario, it's helpful to provide specific examples. When providing examples, consider the following:
 
 - **Describe difficult use cases** where the prompt is ambiguous or complicated, to give the model more visibility into how to approach such cases.
 
@@ -166,7 +166,7 @@ Here are some examples of lines you can include to potentially mitigate differen
 
 Indirect attacks, also referred to as Indirect Prompt Attacks, or Cross Domain Prompt Injection Attacks, are a type of prompt injection technique where malicious instructions are hidden in the ancillary documents that are fed into Generative AI Models. We’ve found system messages to be an effective mitigation for these attacks, by way of spotlighting.
 
-**Spotlighting** is a family of techniques that helps large language models (LLMs) distinguish between valid system instructions and potentially untrustworthy external inputs. It is based on the idea of transforming the input text in a way that makes it more salient to the model, while preserving its semantic content and task performance.
+**Spotlighting** is a family of techniques that helps large language models (LLMs) distinguish between valid system instructions and potentially untrustworthy external inputs. It's based on the idea of transforming the input text in a way that makes it more salient to the model, while preserving its semantic content and task performance.
 
 - **Delimiters** are a natural starting point to help mitigate indirect attacks. Including delimiters in your system message helps to explicitly demarcate the location of the input text in the system message. You can choose one or more special tokens to prepend and append the input text, and the model will be made aware of this boundary. By using delimiters, the model will only handle documents if they contain the appropriate delimiters, which reduces the success rate of indirect attacks. However, since delimiters can be subverted by clever adversaries, we recommend you continue on to the other spotlighting approaches.
 
@@ -182,7 +182,7 @@ Below is an example of a potential system message, for a retail company deployin
 
 :::image type="content" source="../media/concepts/system-message/template.png" alt-text="Screenshot of metaprompts influencing a chatbot conversation." lightbox="../media/concepts/system-message/template.png":::
 
-Finally, remember that system messages, or metaprompts, are not "one size fits all." Use of these type of examples has varying degrees of success in different applications. It is important to try different wording, ordering, and structure of system message text to reduce identified harms, and to test the variations to see what works best for a given scenario.
+Finally, remember that system messages, or metaprompts, are not "one size fits all." Use of these type of examples has varying degrees of success in different applications. It's important to try different wording, ordering, and structure of system message text to reduce identified harms, and to test the variations to see what works best for a given scenario.
 
 ## Next steps
 

@@ -5,7 +5,7 @@ description: Learn about performance and latency with Azure OpenAI
 manager: nitinme
 ms.service: azure-ai-openai
 ms.topic: how-to
-ms.date: 02/07/2024
+ms.date: 09/20/2024
 author: mrbullwinkle 
 ms.author: mbullwin
 recommendations: false
@@ -52,7 +52,7 @@ There are several factors that you can control to improve per-call latency of yo
 
 ### Model selection
 
-Latency varies based on what model you're using. For an identical request, expect that different models have different latencies for the chat completions call. If your use case requires the lowest latency models with the fastest response times, we recommend the latest models in the [GPT-3.5 Turbo model series](../concepts/models.md#gpt-35-models).
+Latency varies based on what model you're using. For an identical request, expect that different models have different latencies for the chat completions call. If your use case requires the lowest latency models with the fastest response times, we recommend the latest [GPT-4o mini model](../concepts/models.md).
 
 ### Generation size and Max tokens
 
@@ -128,7 +128,7 @@ Time from the first token to the last token, divided by the number of generated
 
 ## Summary
 
-* **Model latency**: If model latency is important to you, we recommend trying out our latest models in the [GPT-3.5 Turbo model series](../concepts/models.md).
+* **Model latency**: If model latency is important to you, we recommend trying out the [GPT-4o mini model](../concepts/models.md).
 
 * **Lower max tokens**: OpenAI has found that even in cases where the total number of tokens generated is similar the request with the higher value set for the max token parameter will have more latency.
 

@@ -7,13 +7,13 @@ ms.author: mbullwin
 ms.service: azure-ai-openai
 ms.custom: devx-track-python
 ms.topic: how-to
-ms.date: 02/26/2024
+ms.date: 09/26/2024
 manager: nitinme
 ---
 
 # Migrating to the OpenAI Python API library 1.x
 
-OpenAI has just released a new version of the [OpenAI Python API library](https://github.com/openai/openai-python/). This guide is supplemental to [OpenAI's migration guide](https://github.com/openai/openai-python/discussions/742) and will help bring you up to speed on the changes specific to Azure OpenAI.
+OpenAI released a new version of the [OpenAI Python API library](https://github.com/openai/openai-python/). This guide is supplemental to [OpenAI's migration guide](https://github.com/openai/openai-python/discussions/742) and will help bring you up to speed on the changes specific to Azure OpenAI.
 
 ## Updates
 

@@ -6,9 +6,9 @@ manager: nitinme
 ms.service: azure-ai-openai
 ms.custom: openai
 ms.topic: how-to
-author: ChrisHMSFT
-ms.author: chrhoder
-ms.date: 08/23/2024
+author: mrbullwinkle
+ms.author: mbullwin
+ms.date: 09/20/2024
 recommendations: false
 ---
 

@@ -6,7 +6,7 @@ services: cognitive-services
 manager: nitinme
 ms.service: azure-ai-openai
 ms.topic: how-to
-ms.date: 07/19/2024
+ms.date: 09/20/2024
 author: mrbullwinkle
 ms.author: mbullwin
 recommendations: false
@@ -15,7 +15,7 @@ recommendations: false
 
 # Learn how to use reproducible output (preview)
 
-By default if you ask an Azure OpenAI Chat Completion model the same question multiple times you're likely to get a different response. The responses are therefore considered to be non-deterministic. Reproducible output is a new  preview feature that allows you to selectively change the default behavior to help product more deterministic outputs.
+By default if you ask an Azure OpenAI Chat Completion model the same question multiple times you're likely to get a different response. The responses are therefore considered to be nondeterministic. Reproducible output is a new  preview feature that allows you to selectively change the default behavior to help product more deterministic outputs.
 
 ## Reproducible output support
 
@@ -28,6 +28,7 @@ Reproducible output is only currently supported with the following:
 * `gpt-4` (1106-Preview)
 * `gpt-4` (0125-Preview)
 * `gpt-4` (turbo-2024-04-09)
+* `gpt-4o-mini` (2024-07-18)
 * `gpt-4o` (2024-05-13)
 
 Consult the [models page](../concepts/models.md) for the latest information on model regional availability.

@@ -6,15 +6,15 @@ description: Learn how to use the Codex models on Azure OpenAI to handle a varie
 manager: nitinme
 ms.service: azure-ai-openai
 ms.topic: how-to
-ms.date: 02/15/2024
+ms.date: 09/20/2024
 author: mrbullwinkle
 ms.author: mbullwin
 ---
 
 # Codex models and Azure OpenAI Service
 
-> [!NOTE]
-> This article was authored and tested against the [legacy code generation models](/azure/ai-services/openai/concepts/legacy-models). These models use the completions API, and its prompt/completion style of interaction. If you wish to test the techniques described in this article verbatim we recommend using the `gpt-35-turbo-instruct` model which allows access to the completions API. However, for code generation the chat completions API and the latest GPT-4 models will generally yield the best results, but the prompts would need to be converted to the conversational style specific to interacting with those models.
+> [!IMPORTANT]
+> This article was authored and tested against the [legacy code generation models](/azure/ai-services/openai/concepts/legacy-models). These models use the completions API, and its prompt/completion style of interaction. If you wish to test the techniques described in this article verbatim we recommend using the `gpt-35-turbo-instruct` model which allows access to the completions API. However, for code generation the chat completions API and the latest GPT-4o models will yield the best results, but the prompts would need to be converted to the conversational style specific to interacting with those models.
 
 The Codex model series is a descendant of our GPT-3 series that's been trained on both natural language and billions of lines of code. It's most capable in Python and proficient in over a dozen languages including C#, JavaScript, Go, Perl, PHP, Ruby, Swift, TypeScript, SQL, and even Shell.
 
@@ -26,7 +26,7 @@ You can use Codex for a variety of tasks including:
 - Add comments
 - Rewrite code for efficiency
 
-## How to use the Codex models
+## How to use completions models with code
 
 Here are a few examples of using Codex that can be tested in [Azure OpenAI Studio's](https://oai.azure.com) playground with a deployment of a Codex series model, such as `code-davinci-002`.
 

@@ -1,5 +1,5 @@
 ---
-title: 'Quickstart: Use GPT-35-Turbo via the Azure OpenAI Studio'
+title: 'Quickstart: Use GPT-4o and GPT-4o mini via the Azure OpenAI Studio'
 titleSuffix: Azure OpenAI Service
 description: Walkthrough on how to get started with Azure OpenAI and make your first completions call with Azure OpenAI Studio. 
 #services: cognitive-services

@@ -5,7 +5,7 @@ description: Learn how to use Azure OpenAI on your Azure Search data Python & RE
 manager: nitinme
 ms.service: azure-ai-openai
 ms.topic: conceptual
-ms.date: 03/12/2024
+ms.date: 09/20/2024
 author: mrbullwinkle
 ms.author: mbullwin
 recommendations: false