Are Azure OpenAI models identical to their equivalent models in OpenAI

Question

When comparing the same revision/static models for a model in OpenAI vs Azure OpenAI are the models identical? If we wanted to mix and match traffic to both providers is it safe to assume that responses will be the same from either provider assuming we specify the exact static model (i.e. don't utilize auto-upgrade models).

Answer 1

@David Zhai
Thanks for reaching out to us. I have seen your other post, but still answer on this side in case you are not aware of the other one. Please keep in mind that if you're using the model's random generation capabilities (for example, by setting a temperature parameter higher than 0), then the model will introduce randomness into its outputs, and you might get different results for the same input. Even for the same model in the same platform with the same input, the output is not the same if you have a temperature higher than 0. If you're using different "tokens" or "prompts" as input, even if the text is the same, the output might be different. This is because the model considers the entire input, including prompts, when generating output. parameters are below screenshot, given Azure OpenAI as an example - To answer the question generally, OpenAI and Microsoft have a partnership, and as part of this, Microsoft hosts OpenAI's models on Azure. However, while they strive to maintain consistency between the models, there may be differences due to various factors such as updates, underlying infrastructure, or software versions. Using a specific static model, such as gpt-3.5-turbo-1106, should in theory provide the same results across both platforms for the same input with no random generation capabilities, as long as the model version is identical, and no upgrades have been applied. However, it's important to note that even with the same model, the results can sometimes have subtle differences due to the inherent randomness in these AI models. Also, the performance, latency, and other factors may vary due to different underlying infrastructure and configurations. To ensure consistency and accurate results, it's recommended to stick with one platform for all related requests. If you plan to split traffic between OpenAI and Azure, it would be good to thoroughly test the system to ensure it meets your requirements. Please let us know if your scenario is specific, we can discuss with product team to see if there any solution to stick with one platform to avoid the concern. I hope this helps. Regards, Yutong -Please kindly accept the answer if you feel helpful to support the community, thanks a lot.