Code not formatted correctly with Azure OpenAI o3-mini

[rubenaster]

What happened?

The code-view for user and assistant messages is falsly formatted.

This is my user prompt:

Do not make this calling a module but write the resources directly, instead.

"""
module "openai" {
  source = "./azure-openai"

  name = local.openai_name

  resource_group_name = data.azurerm_resource_group.main.name
  location            = "swedencentral"

  public_network_access_enabled = false
  local_auth_enabled            = true

  deployments = var.openai.deployments

  private_endpoints = [{
    name                            = "${local.openai_name}-pe"
    private_service_connection_name = "${local.openai_name}-connection"

    location  = azurerm_virtual_network.main.location
    vnet_id   = azurerm_virtual_network.main.id
    subnet_id = azurerm_subnet.private_endpoints.id

    private_dns_zone_name = module.openai_private_dns.zone_name
    private_dns_zone_id   = module.openai_private_dns.zone_id

    private_dns_entry_enabled          = true
    dns_zone_virtual_network_link_name = "dns_zone_link"
    is_manual_connection               = false
  }]

  tags = local.base_tags
}
"""

This is what the user prompt UI looks like (HCL code):

I have similar issues with the assistant response. Sometimes the formatting issues look the same, sometimes code isn't formatted at all, like on the following screenshot (HCL code):

Here's another example with code being rendered in the middle of the code for the assistant output (Python code):

---

Choosing for example Azure GPT-4o-mini doesn't render the user message correct but at least renders the assistant output correct.

Version Information

I'm using Azure App Service with Docker:

librechat/librechat:v0.7.7
from https://index.docker.io

OpenAI API-version '2025-02-01-preview'
o3-mini version: '2025-01-31'

Steps to Reproduce

1. Deploy LibreChat v0.7.7 with Azure OpenAI o3-mini configured
2. Select o3-mini as Model
3. Write the prompt mentioned above and observe the output

What browsers are you seeing the problem on?

Microsoft Edge

Relevant log output

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

There is discussion on this already, please search existing discussions.

https://platform.openai.com/docs/guides/reasoning-best-practices

[danny-avila]

Read the article I linked, this is how the model was trained to output text unless you give it a very specific prompt.

[danny-avila]

This section specifically https://platform.openai.com/docs/guides/reasoning-best-practices#how-to-prompt-reasoning-models-effectively

[rubenaster]

I understand. But that the user message is incorrectly formatted shouldn't relate to the matter described in the article.

[danny-avila]

Using triple-quotation marks """ is a python format, not markdown. Use triple-backticks ``` instead

[rubenaster]

Triple quotes should not format code but are usually being used to encapsule sections within larger prompts.

But I see that one needs to adjust its behaviour when switching from ChatGPT to LibreChat.