Azure OpenAI - Assistants Bugs

Problem 1: Inconsistent Function Saving

  • Issue: The Assistants Playground does not consistently save functions. There are two playgrounds: one from Azure OAI Studio and one from Azure AI Studio.
  • Details: Despite trying both playgrounds, functions are often missing upon logging back in.

Problem 2: Performance and Availability Issues

  • Issue: Inconsistent performance and availability of models on the Azure Playground compared to the OpenAI Playground.
  • Details:
    • When an action is required, most GPT-3.5 models fail, resulting in server errors. Screenshots of these issues have been sent to Microsoft, but no response has been received.
    • Adjusting the temperature to 0.1 using the mouse causes the page to crash.
    • Some 3.5 turbo models are not available on the Azure Playground, unlike on the OpenAI Playground. Server changes and tests did not resolve these issues, leading to frustration.
    • Only one GPT-3.5 model reliably follows instructions, while the others are ineffective.
    • The GPT-4 model performs exceptionally well, but the quota limit is reached quickly. Despite having Microsoft Sponsorship, the frequent downtime and quota issues with GPT-3.5 models are problematic.

Problem 3: Code Copying Issues

  • Issue: Copying code for the Assistant in Azure Playground can result in broken code if not properly formatted.
  • Details: To prevent this, include the parameter instructions="""You are a nice Assistant...""" within the docstrings to ensure the code functions correctly. This is crucial for new users who might encounter issues with improperly formatted copied code.


I recommend that you flag these issues directly with the Azure support or community. This Forum can’t really assist with or address Azure OpenAI related matters.

I haven’t used it myself, but I believe this would be the Community for such matters:


Hi Jen!
Could you please provide more information about who supports the models on the Microsoft platform? Additionally, any relevant links would be appreciated, as I am unsure about the current situation. I do not want to build on older models and wait indefinitely for improvements.

I’m afraid I don’t have visibility into that.

As per above, I suggest looking through the existing discussions in the Azure community Q&A (filter by OpenAI) and/or submitting your own question. There seems to be a decent level of activity and engagement there:

Additionally, this page offers updates on Azure OpenAI:

Perhaps others have additional input.