Process for Azure OpenAI Signup

Given some of the API performance/stability issues recently and the supposedly better performance with Azure APIs I thought I’d try signing up for Azure OpenAI APIs. Sharing here for anyone else who isn’t already in the Azure ecosystem and is curious, knowing that some of these Enterprise services can be a real challenge to get started.
(not affiliate with Azure or OpenAI)

  1. Head to Azure OpenAI page and Start Free and signup
  2. Once your in Azure Portal, Add Product and search for OpenAI
  3. You’ll get a link to the Request Access to Azure OpenAI Service form. Some notes:
    1. Must use a company email. “Applications submitted with a personal email address (e.g. gmail.com, outlook.com, etc.) will be DENIED.”
    2. Must apply on behalf of your own company. Not your client’s company
    3. Agree to Azure Data Policy and Code of Conduct for the Azure OpenAI Service
  4. Once approved (took 1 business day), head to Azure AI Studio and setup “AI Resource”
    1. Be sure to save the approval email. You’ll need it for GPT4
  5. Create a Deployment of the model(s) you want to use
  6. Then you can head to Playground and test it out. You’ll get a unique Base URL and API Key for each Deployment/model
    • Like: https://example-company.openai.azure.com/openai/deployments/gpt-35-turbo/chat/completions?api-version=2023-03-15-preview
    • Playground is pretty similar
      • Gives you option to copy your conversation as python/js/c#/curl
      • Has some example presets of different System Messages, and lets you enter in few-shot examples in the form
      • Has a nice token breakdown display
  7. Then you can apply for GPT4. (still waiting on this one, been 2 weeks)

Overall pretty straightforward, initial approval was fast, no weird “Enterprise” requirements, but seems like you do need a company, not intended for personal use.

3 Likes

And what I do as an open-source developer who can’t provide company details?

2 Likes

This is the reason why many people don’t like microsoft, Just use OpenAI API, I’m sure will be much more stable soon, send requests from eastern part of US like Arizona, Ohio, and Atlanta … This will help to have lower latency, Or try using streaming. The new GPT-4 model is much more faster, more cleaner language.

Thanks so much for the step by step walkthrough. If we use Azure for GPT4 are we able to spend more than $120/month without requesting an increase?