Commercial usage of fine-tuning a model using synthetic data generated by Azure Open AI models

Question

Commercial usage of fine-tuning a model using synthetic data generated by Azure Open AI models

II 80

Hi,

Is it permissible for me to generate synthetic data using Azure Open AI models (such as o4-mini) and later use that synthetically generated data to fine-tune a model and use it for commercial purposes?

I know its not possible if I use Open AI directly (https://community.openai.com/t/synthetic-instructions-generated-by-openai/569888), I would like to know if that is permissible through Azure Open AI.

Also I would like to know which other models allow me to do this via Azure AI Foundry (eg: DeepSeek, Mistral, etc)?

Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-26T11:29:52.1766667+00:00

Hi II

Yes, you can use the chat completion call to finetune your model in Azure OpenAI under Distillation feature (Preview)

Available for Azure OpenAI models mostly from UI. You can probably download the data using SDK for finetuning against other supported models

Attached relevant document for reference.

https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/stored-completions?tabs=python-secure#distillation

Thank you
II 80 Reputation points

2025-08-26T11:59:27.9433333+00:00

Hi @Manas Mohanty ,

Thanks for the answer - I am not looking for distillation process here, rather I am looking regarding, synthetic data generation. For example, I ask the model to "Generate 5 patient questions and answers related to managing hypertension, suitable for training a medical assistant chatbot." via API and later use the response from the model to fine-tune another model (open-source LLM to be used commercially), that is different then distillation. Is that allowed?
Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-27T01:31:32.93+00:00

Hi II

Finetuning operation ask Jsonl input which can be synthetically generated by models themselves.

https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning?tabs=azure-openai&pivots=programming-language-studio#prepare-your-training-and-validation-data

But let me check any ethical side of issue with team internally and get back to you.

Thank you
Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-28T01:44:08.02+00:00

Hi II

No limitation confirmed from product group on finetuning with generated data as we suspected.

Converted my first comment answer for acceptance.

Thank you.
II 80 Reputation points

2025-08-28T08:57:29.0933333+00:00
Hi again,

Thanks for your effort, however you're communication is still not clear to us.

To provide more context, in Open AI's Terms of Use, they mentioned "Use Output to develop models that compete with OpenAI." is prohibited. This can include synthetic data generated through Open AI models as well.

Our questions are:

We would like to know, is the same limitation applicable in Azure Open AI models too?

We're looking for an answer in this context to know if it is permissible to generate synthetic data via API calls to Azure Open AI model to fine-tune a different model for commercial usage?

(ie) If I use an Azure Open AI model like o4-mini -> generate new synthetic data from it -> download the synthetic data locally -> fine-tune a locally hosted, open-source model on this synthetic data -> use this model commercially. Is this allowed?

Please could you get an idea for these questions 1, 2 and (2.a) from PG and update us?
Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-29T06:30:07.19+00:00
Hi II

Here is confirmation from product group.

We would like to know, is the same limitation applicable in Azure Open AI models too? - Not applicable for Azure OpenAI

We're looking for an answer in this context to know if it is permissible to generate synthetic data via API calls to Azure Open AI model to fine-tune a different model for commercial usage?

(ie) If I use an Azure Open AI model like o4-mini -> generate new synthetic data from it -> download the synthetic data locally -> fine-tune a locally hosted, open-source model on this synthetic data -> use this model commercially. Is this allowed? - Yes, confirmed that no restrictions on Models directly hosted from Azure.
II 80 Reputation points

2025-08-29T08:49:58.32+00:00

Thanks. Please could you modify your original answer, so I can accept the modified one as answer

Accepted answer

0 additional answers

Your answer

Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-26T11:29:52.1766667+00:00

Hi II

Yes, you can use the chat completion call to finetune your model in Azure OpenAI under Distillation feature (Preview)

Available for Azure OpenAI models mostly from UI. You can probably download the data using SDK for finetuning against other supported models

Attached relevant document for reference.

https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/stored-completions?tabs=python-secure#distillation

Thank you
II 80 Reputation points

2025-08-26T11:59:27.9433333+00:00

Hi @Manas Mohanty ,

Thanks for the answer - I am not looking for distillation process here, rather I am looking regarding, synthetic data generation. For example, I ask the model to "Generate 5 patient questions and answers related to managing hypertension, suitable for training a medical assistant chatbot." via API and later use the response from the model to fine-tune another model (open-source LLM to be used commercially), that is different then distillation. Is that allowed?
Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-27T01:31:32.93+00:00

Hi II

Finetuning operation ask Jsonl input which can be synthetically generated by models themselves.

https://learn.microsoft.com/en-us/azure/ai-foundry/openai/how-to/fine-tuning?tabs=azure-openai&pivots=programming-language-studio#prepare-your-training-and-validation-data

But let me check any ethical side of issue with team internally and get back to you.

Thank you
Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-28T01:44:08.02+00:00

Hi II

No limitation confirmed from product group on finetuning with generated data as we suspected.

Converted my first comment answer for acceptance.

Thank you.
II 80 Reputation points

2025-08-28T08:57:29.0933333+00:00

Hi again,

Thanks for your effort, however you're communication is still not clear to us.

To provide more context, in Open AI's Terms of Use, they mentioned "Use Output to develop models that compete with OpenAI." is prohibited. This can include synthetic data generated through Open AI models as well.

Our questions are:

We would like to know, is the same limitation applicable in Azure Open AI models too?

We're looking for an answer in this context to know if it is permissible to generate synthetic data via API calls to Azure Open AI model to fine-tune a different model for commercial usage?

(ie) If I use an Azure Open AI model like o4-mini -> generate new synthetic data from it -> download the synthetic data locally -> fine-tune a locally hosted, open-source model on this synthetic data -> use this model commercially. Is this allowed?

Please could you get an idea for these questions 1, 2 and (2.a) from PG and update us?
Manas Mohanty 9,655 Reputation points Microsoft External Staff Moderator

2025-08-29T06:30:07.19+00:00

Hi II

Here is confirmation from product group.

We would like to know, is the same limitation applicable in Azure Open AI models too? - Not applicable for Azure OpenAI

We're looking for an answer in this context to know if it is permissible to generate synthetic data via API calls to Azure Open AI model to fine-tune a different model for commercial usage?

(ie) If I use an Azure Open AI model like o4-mini -> generate new synthetic data from it -> download the synthetic data locally -> fine-tune a locally hosted, open-source model on this synthetic data -> use this model commercially. Is this allowed? - Yes, confirmed that no restrictions on Models directly hosted from Azure.
II 80 Reputation points

2025-08-29T08:49:58.32+00:00

Thanks. Please could you modify your original answer, so I can accept the modified one as answer

Answer 1

Manas Mohanty 9,655 Microsoft External Staff Moderator

Hi II

It is confirmed from product group that you can finetune Azure OpenAI models with Chat completions created by Azure OpenAI model (Same restriction of OpenAI does not apply).

In fact, we have distillation process to fine tune models with chat completion call.

You can also the JSONL containing the Synthetic data to finetune with other Azure Hosted models too.

Thank you.

Share via

Commercial usage of fine-tuning a model using synthetic data generated by Azure Open AI models

0 additional answers

Your answer